1<!DOCTYPE html><html lang="en-US"> <head><meta charset="utf-8"><meta http-equiv="X-UA-Compatible" content="IE=edge"><meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes"><meta name="generator" content="Astro v5.15.9"><meta name="robots" content="index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1"><meta name="googlebot" content="index, follow"><meta name="google" content="notranslate"><meta name="revisit-after" content="1 days"><!-- SEO Enhancement Tags --><meta name="author" content="Farez Vadsaria"><meta name="publisher" content="Farez Vadsaria"><meta name="language" content="en-US"><meta name="geo.region" content="CA"><meta name="geo.placename" content="Canada"><!-- Mobile and PWA Tags --><meta name="mobile-web-app-capable" content="yes"><meta name="apple-mobile-web-app-capable" content="yes"><meta name="apple-mobile-web-app-status-bar-style" content="default"><meta name="apple-mobile-web-app-title" content="Farez Vadsaria"><meta name="format-detection" content="telephone=no,date=no,address=no,email=no,url=no"><!-- Theme Colors --><meta name="theme-color" content="#121212" media="(prefers-color-scheme: dark)"><meta name="theme-color" content="#ffffff" media="(prefers-color-scheme: light)"><meta name="msapplication-TileColor" content="#121212"><!-- Feed link --><link rel="manifest" href="/site.webmanifest"><link rel="alternate" type="application/rss+xml" title="Farez Vadsaria RSS Feed" href="https://www.farez.ai/rss.xml"><link rel="icon" type="image/png" href="/favicon-96x96.png" sizes="96x96"><link rel="icon" type="image/svg+xml" href="/favicon.svg"><link rel="shortcut icon" href="/favicon.ico"><link rel="apple-touch-icon" sizes="180x180" href="/apple-touch-icon.png"><meta name="apple-mobile-web-app-title" content="Farez Vadsaria"><link rel="manifest" href="/site.webmanifest"><title>The Field Guide: a map of AI concepts | Farez Vadsaria</title><meta name="description" content="The Field Guide: 549 AI and machine-learning terms on one map. Pick any destination and follow a route through exactly the ideas you need first."><!-- robots set globally in base head to avoid duplicates --><meta name="language" content="en-US"><link rel="canonical" href="https://www.farez.ai/guide/"><!-- Open Graph Meta Tags --><meta property="og:title" content="The Field Guide: a map of AI concepts"><meta property="og:description" content="The Field Guide: 549 AI and machine-learning terms on one map. Pick any destination and follow a route through exactly the ideas you need first."><meta property="og:image" content="https://www.farez.ai/og-default.png"><meta property="og:image:alt" content="The Field Guide: a map of AI concepts"><meta property="og:image:width" content="1200"><meta property="og:image:height" content="630"><meta property="og:type" content="website"><meta property="og:locale" content="en-US"><meta property="og:site_name" content="Farez Vadsaria"><meta property="og:url" content="https://www.farez.ai/guide/"><!-- Twitter Card Meta Tags --><meta name="twitter:card" content="summary_large_image"><meta name="twitter:title" content="The Field Guide: a map of AI concepts"><meta name="twitter:description" content="The Field Guide: 549 AI and machine-learning terms on one map. Pick any destination and follow a route through exactly the ideas you need first."><meta name="twitter:image" content="https://www.farez.ai/og-default.png"><meta name="twitter:image:alt" content="The Field Guide: a map of AI concepts"><meta name="twitter:creator" content="@farezvadsaria"><meta name="twitter:site" content="@farezvadsaria"><meta name="twitter:domain" content="farez.ai"><!-- Enhanced JSON-LD Structured Data -->
1<script type="application/ld+json">{"@context":"https://schema.org","@type":"WebSite","@id":"https://farez.ai#website","url":"https://farez.ai","name":"Farez Vadsaria","description":"The Field Guide: 549 AI and machine-learning terms on one map. Pick any destination and follow a route through exactly the ideas you need first.","inLanguage":"en-US","publisher":{"@type":"Person","name":"Farez Vadsaria","url":"https://farez.ai"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://farez.ai/blog?q={search_term_string}"},"query-input":"required name=search_term_string"},"mainEntity":{"@type":"Blog","@id":"https://farez.ai/blog#blog","name":"Farez Vadsaria Blog","description":"Technology blog covering web development, programming, and software engineering topics","url":"https://farez.ai/blog","author":{"@type":"Person","name":"Farez Vadsaria"},"blogPost":[{"@type":"BlogPosting","headline":"The Evolution of Developer-Targeted Recruitment Scams: From Fake Profiles to Spoofed Commits","description":"How sophisticated social engineering attacks exploit developer trust through fake LinkedIn profiles, spoofed GitHub commits, and malicious npm packages. Learn how to protect yourself from these multi-layer attacks.","url":"https://farez.ai/blog/how-claude-saved-me-with-targeted-recruitment-scams/","datePublished":"2025-09-22T00:00:00.000Z","author":{"@type":"Person","name":"Farez Vadsaria"},"keywords":"security, social-engineering, github, npm"},{"@type":"BlogPosting","headline":"3 reasons why your business needs an AI voice call bot","description":"Discover 3 reasons your business needs an AI voice call bot powered by LLMs, with practical steps to get started and a quick overview of Hume AI, VAPI, and ElevenLabs.","url":"https://farez.ai/blog/3-reasons-why-your-business-needs-a-voice-bot/","datePublished":"2025-04-11T00:00:00.000Z","author":{"@type":"Person","name":"Farez Vadsaria"},"keywords":"voice-ai, llm, business"}]}}</script>
1<script src="https://analytics.ahrefs.com/analytics.js" data-key="Pdo0jMTPh2TKWfqNr0glyQ" async></script>
1<script>(function(){const openpanelClientId = "33e6d663-ca64-4d5f-acad-fd639bc80962"; 2 3 window.op = window.op || function(...args) { (window.op.q = window.op.q || []).push(args); }; 4 window.op('init', { 5 clientId: openpanelClientId, 6 trackScreenViews: true, 7 trackOutgoingLinks: true, 8 trackAttributes: true, 9 }); 10 })();</script>
10<script src="https://openpanel.dev/op1.js" defer async></script>
10<link rel="preload" href="/fonts/ClashDisplay/ClashDisplay-Semibold.woff2" as="font" type="font/woff2" crossorigin><link rel="preload" href="/fonts/Lexend/Lexend-Regular.woff2" as="font" type="font/woff2" crossorigin><link rel="dns-prefetch" href="//analytics.ahrefs.com"><link rel="dns-prefetch" href="//openpanel.dev">
10<script data-astro-rerun> 11 (function() { 12 try { 13 const localStorageTheme = localStorage.getItem('theme'); 14 const isDark = 15 (localStorageTheme === 'dark') || 16 (!localStorageTheme && window.matchMedia('(prefers-color-scheme: dark)').matches); 17 18 document.documentElement.classList.toggle('dark', isDark); 19 localStorage.setItem('theme', isDark ? 'dark' : 'light'); 20 21 window.matchMedia('(prefers-color-scheme: dark)').addEventListener('change', e => { 22 if (!localStorage.getItem('theme')) { 23 document.documentElement.classList.toggle('dark', e.matches); 24 localStorage.setItem('theme', e.matches ? 'dark' : 'light'); 25 } 26 }); 27 } catch (error) { 28 console.warn('Theme setting failed:', error); 29 } 30 })(); 31 </script>
31<link rel="alternate" type="application/rss+xml" title="RSS Feed" href="/rss.xml"><link rel="stylesheet" href="/_astro/_id_.CY6RqFOK.css"> 32<link rel="stylesheet" href="/_astro/_concept_.UvjIE6e0.css"></head><body> <div class="flex h-fit min-h-screen w-full flex-col gap-y-4 sm:gap-y-6 font-sans"> <style>astro-island,astro-slot,astro-static-slot{display:contents}</style>
32<script>(()=>{var e=async t=>{await(await t())()};(self.Astro||(self.Astro={})).load=e;window.dispatchEvent(new Event("astro:load"));})();</script>
32<script>(()=>{var A=Object.defineProperty;var g=(i,o,a)=>o in i?A(i,o,{enumerable:!0,configurable:!0,writable:!0,value:a}):i[o]=a;var d=(i,o,a)=>g(i,typeof o!="symbol"?o+"":o,a);{let i={0:t=>m(t),1:t=>a(t),2:t=>new RegExp(t),3:t=>new Date(t),4:t=>new Map(a(t)),5:t=>new Set(a(t)),6:t=>BigInt(t),7:t=>new URL(t),8:t=>new Uint8Array(t),9:t=>new Uint16Array(t),10:t=>new Uint32Array(t),11:t=>1/0*t},o=t=>{let[l,e]=t;return l in i?i[l](e):void 0},a=t=>t.map(o),m=t=>typeof t!="object"||t===null?t:Object.fromEntries(Object.entries(t).map(([l,e])=>[l,o(e)]));class y extends HTMLElement{constructor(){super(...arguments);d(this,"Component");d(this,"hydrator");d(this,"hydrate",async()=>{var b;if(!this.hydrator||!this.isConnected)return;let e=(b=this.parentElement)==null?void 0:b.closest("astro-island[ssr]");if(e){e.addEventListener("astro:hydrate",this.hydrate,{once:!0});return}let c=this.querySelectorAll("astro-slot"),n={},h=this.querySelectorAll("template[data-astro-template]");for(let r of h){let s=r.closest(this.tagName);s!=null&&s.isSameNode(this)&&(n[r.getAttribute("data-astro-template")||"default"]=r.innerHTML,r.remove())}for(let r of c){let s=r.closest(this.tagName);s!=null&&s.isSameNode(this)&&(n[r.getAttribute("name")||"default"]=r.innerHTML)}let p;try{p=this.hasAttribute("props")?m(JSON.parse(this.getAttribute("props"))):{}}catch(r){let s=this.getAttribute("component-url")||"<unknown>",v=this.getAttribute("component-export");throw v&&(s+=` (export ${v})`),console.error(`[hydrate] Error parsing props for component ${s}`,this.getAttribute("props"),r),r}let u;await this.hydrator(this)(this.Component,p,n,{client:this.getAttribute("client")}),this.removeAttribute("ssr"),this.dispatchEvent(new CustomEvent("astro:hydrate"))});d(this,"unmount",()=>{this.isConnected||this.dispatchEvent(new CustomEvent("astro:unmount"))})}disconnectedCallback(){document.removeEventListener("astro:after-swap",this.unmount),document.addEventListener("astro:after-swap",this.unmount,{once:!0})}connectedCallback(){if(!this.hasAttribute("await-children")||document.readyState==="interactive"||document.readyState==="complete")this.childrenConnectedCallback();else{let e=()=>{document.removeEventListener("DOMContentLoaded",e),c.disconnect(),this.childrenConnectedCallback()},c=new MutationObserver(()=>{var n;((n=this.lastChild)==null?void 0:n.nodeType)===Node.COMMENT_NODE&&this.lastChild.nodeValue==="astro:end"&&(this.lastChild.remove(),e())});c.observe(this,{childList:!0}),document.addEventListener("DOMContentLoaded",e)}}async childrenConnectedCallback(){let e=this.getAttribute("before-hydration-url");e&&await import(e),this.start()}async start(){let e=JSON.parse(this.getAttribute("opts")),c=this.getAttribute("client");if(Astro[c]===void 0){window.addEventListener(`astro:${c}`,()=>this.start(),{once:!0});return}try{await Astro[c](async()=>{let n=this.getAttribute("renderer-url"),[h,{default:p}]=await Promise.all([import(this.getAttribute("component-url")),n?import(n):()=>()=>{}]),u=this.getAttribute("component-export")||"default";if(!u.includes("."))this.Component=h[u];else{this.Component=h;for(let f of u.split("."))this.Component=this.Component[f]}return this.hydrator=p,this.hydrate},e,this)}catch(n){console.error(`[astro-island] Error hydrating ${this.getAttribute("component-url")}`,n)}}attributeChangedCallback(){this.hydrate()}}d(y,"observedAttributes",["props"]),customElements.get("astro-island")||customElements.define("astro-island",y)}})();</script>
32<astro-island uid="zbqvc" prefix="r2" component-url="/_astro/navbar.DrvGXZjr.js" component-export="default" renderer-url="/_astro/client.Z6S47nf_.js" props="{}" ssr client="load" opts="{"name":"Navbar","value":true}" await-children><header aria-label="Navigation" role="navigation" class="fixed left-1/2 z-30 -translate-x-1/2 transform backdrop-blur-lg bg-background/80 rounded-none shadow-none transition-all duration-300 ease-in-out border border-transparent w-full top-2 lg:top-4 xl:top-6" style="width:100%"><div class="mx-auto flex max-w-7xl items-center justify-between gap-4 p-4"><a href="/" target="_self" class="transition-colors duration-300 ease-in-out font-custom flex shrink-0 items-center gap-2 text-xl font-bold" aria-label="Home" title="Home" navigation="true"><svg viewBox="0 0 350 266" xmlns="http://www.w3.org/2000/svg" class="text-neutral-900 dark:text-neutral-100 transition-colors duration-300 h-8 w-8" role="img" aria-label="Logo" aria-hidden="true" name="logo" focusable="false"><path d="M0 0h216.667C290.305 0 350 59.546 350 133s-59.695 133-133.333 133L0 0zm231.629 231.641c48.131-7.203 85.038-48.623 85.038-98.641 0-55.09-44.772-99.75-100-99.75h-54.862C185.557 59.722 200 94.678 200 133c0 17.331-2.96 33.991-8.405 49.492l40.034 49.149zm-66.243-81.326c.844-5.646 1.281-11.428 1.281-17.315 0-42.333-22.66-79.386-56.541-99.75H70.032l95.354 117.065z" fill="currentColor" fill-rule="evenodd" clip-rule="evenodd"></path></svg><span class="transition-opacity duration-200 ease-in-out text-foreground/90 dark:text-white">Farez Vadsaria</span></a><div class="flex items-center gap-2 md:gap-4"><nav class="hidden items-center gap-6 md:flex" aria-label="Main navigation" role="navigation"><div class="relative"><a href="/learn" target="_self" class="inline-block ease-in-out text-sm font-medium capitalize transition-colors duration-200 relative py-1 px-1 after:absolute after:bottom-0 after:left-0 after:h-[2px] after:w-0 after:bg-primary after:transition-all after:duration-300 hover:after:w-full hover:text-foreground text-foreground/70">learn</a></div><div class="relative"><a href="/guide" target="_self" class="inline-block ease-in-out text-sm font-medium capitalize transition-colors duration-200 relative py-1 px-1 after:absolute after:bottom-0 after:left-0 after:h-[2px] after:w-0 after:bg-primary after:transition-all after:duration-300 hover:after:w-full hover:text-foreground text-foreground/70">field guide</a></div><div class="relative"><a href="/claude-guide" target="_self" class="inline-block ease-in-out text-sm font-medium capitalize transition-colors duration-200 relative py-1 px-1 after:absolute after:bottom-0 after:left-0 after:h-[2px] after:w-0 after:bg-primary after:transition-all after:duration-300 hover:after:w-full hover:text-foreground text-foreground/70">claude guide</a></div><div class="relative"><a href="/harnesses" target="_self" class="inline-block ease-in-out text-sm font-medium capitalize transition-colors duration-200 relative py-1 px-1 after:absolute after:bottom-0 after:left-0 after:h-[2px] after:w-0 after:bg-primary after:transition-all after:duration-300 hover:after:w-full hover:text-foreground text-foreground/70">harnesses</a></div><div class="relative"><a href="/blog" target="_self" class="inline-block ease-in-out text-sm font-medium capitalize transition-colors duration-200 relative py-1 px-1 after:absolute after:bottom-0 after:left-0 after:h-[2px] after:w-0 after:bg-primary after:transition-all after:duration-300 hover:after:w-full hover:text-foreground text-foreground/70">blog</a></div><div class="relative"><a href="/projects" target="_self" class="inline-block ease-in-out text-sm font-medium capitalize transition-colors duration-200 relative py-1 px-1 after:absolute after:bottom-0 after:left-0 after:h-[2px] after:w-0 after:bg-primary after:transition-all after:duration-300 hover:after:w-full hover:text-foreground text-foreground/70">projects</a></div><div tabindex="0"><a href="#waitlist" target="_self" class="ease-in-out bg-primary text-primary-foreground hover:bg-primary/90 inline-flex items-center justify-center rounded-md px-4 py-2 text-sm font-semibold transition-colors duration-200 shadow-sm">Subscribe to the newsletter</a></div></nav><button data-slot="button" class="inline-flex items-center justify-center gap-2 whitespace-nowrap rounded-full text-sm font-medium transition-all disabled:pointer-events-none disabled:opacity-50 [&_svg]:pointer-events-none [&_svg:not([class*='
32size-'])]:size-4 shrink-0 [&_svg]:shrink-0 outline-none focus-visible:border-ring focus-visible:ring-ring/50 focus-visible:ring-[3px] aria-invalid:ring-destructive/20 dark:aria-invalid:ring-destructive/40 aria-invalid:border-destructive bg-secondary text-secondary-foreground shadow-xs hover:bg-secondary/80 size-9" id="theme-toggle" title="Toggle theme"><svg xmlns="http://www.w3.org/2000/svg" width="24" height="24" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-sun size-4 scale-100 rotate-0 transition-all dark:scale-0 dark:-rotate-90" aria-hidden="true"><circle cx="12" cy="12" r="4"></circle><path d="M12 2v2"></path><path d="M12 20v2"></path><path d="m4.93 4.93 1.41 1.41"></path><path d="m17.66 17.66 1.41 1.41"></path><path d="M2 12h2"></path><path d="M20 12h2"></path><path d="m6.34 17.66-1.41 1.41"></path><path d="m19.07 4.93-1.41 1.41"></path></svg><svg xmlns="http://www.w3.org/2000/svg" width="24" height="24" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-moon absolute size-4 scale-0 rotate-90 transition-all dark:scale-100 dark:rotate-0" aria-hidden="true"><path d="M20.985 12.486a9 9 0 1 1-9.473-9.472c.405-.022.617.46.402.803a6 6 0 0 0 8.268 8.268c.344-.215.825-.004.803.401"></path></svg><span class="sr-only">Toggle theme</span></button></div></div></header><!--astro:end--></astro-island> <main class="flex grow flex-col bg-background mt-16 lg:mt-6" aria-label="Main content" role="main"> <div class="mx-auto flex grow flex-col w-full"> <section class="fg-band"> <div class="fg-band__inner"> <p class="fg-mono">The Field Guide · 549 terms · 8 regions · 62 landmarks</p> <h1 class="fg-display mt-3 text-5xl sm:text-6xl">Field Guide</h1> <p class="fg-dim mt-4 max-w-2xl text-lg"> 33A map of AI, not a syllabus. Pick any destination and follow a route through exactly the ideas you need first. 34</p> <div class="fg-search mt-8" data-fg-search> <svg class="fg-search__icon" width="18" height="18" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" aria-hidden="true"><circle cx="11" cy="11" r="7"></circle><path d="m20 20-3.5-3.5"></path></svg> <input type="search" placeholder="Where do you want to go? Try âRLHFâ or âattentionâ" aria-label="Find a destination in the Field Guide" aria-controls="fg-results" aria-autocomplete="list" autocomplete="off" data-fg-input> <span class="fg-search__kbd" aria-hidden="true">/</span> <ul id="fg-results" class="fg-results" role="listbox" hidden data-fg-results></ul> </div> <div class="fg-map-wrap mt-10"><svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 1200 720" class="fg-map" role="img" aria-label="Map of the Field Guide: 549 terms across 8 regions, with 62 landmark terms labelled"><defs><radialGradient id="fg-foundations"><stop offset="0" stop-color="#d9c7a1" stop-opacity=".16"/><stop offset=".75" stop-color="#d9c7a1" stop-opacity=".05"/><stop offset="1" stop-color="#d9c7a1" stop-opacity="0"/></radialGradient><radialGradient id="fg-neural-nets"><stop offset="0" stop-color="#ff6b35" stop-opacity=".16"/><stop offset=".75" stop-color="#ff6b35" stop-opacity=".05"/><stop offset="1" stop-color="#ff6b35" stop-opacity="0"/></radialGradient><radialGradient id="fg-training"><stop offset="0" stop-color="#ffc857" stop-opacity=".16"/><stop offset=".75" stop-color="#ffc857" stop-opacity=".05"/><stop offset="1" stop-color="#ffc857" stop-opacity="0"/></radialGradient><radialGradient id="fg-evaluation"><stop offset="0" stop-color="#9ad97a" stop-opacity=".16"/><stop offset=".75" stop-color="#9ad97a" stop-opacity=".05"/><stop offset="1" stop-color="#9ad97a" stop-opacity="0"/></radialGradient><radialGradient id="fg-language"><stop offset="0" stop-color="#3de0c5" stop-opacity=".16"/><stop offset=".75" stop-color="#3de0c5" stop-opacity=".05"/><stop offset="1" stop-color="#3de0c5" stop-opacity="0"/></radialGradient><radialGradient id="fg-vision"><stop offset="0" stop-color="#5aa9ff" stop-opacity=".16"/><stop offset=".75" stop-color="#5aa9ff" stop-opacity=".05"/><stop offset="1" stop-color="#5aa9ff" stop-opacity="0"/></radialGradient><radialGradient id="fg-agents"><stop offset="0" stop-color="#7b6cff" stop-opacity=".16"/><stop offset=".75" stop-color="#7b6cff" stop-opacity=".05"/><stop offset="1" stop-color="#7b6cff" stop-opacity="0"/></radialGradient><radialGradient id="fg-shipping"><stop offset="0" stop-color="#ff5d8f" stop-opacity=".16"/><stop offset=".75" stop-color="#ff5d8f" stop-opacity=".05"/><stop offset="1" stop-color="#ff5d8f" stop-opacity="0"/></radialGradient></defs><ellipse cx="175" cy="190" rx="146.9" ry="117.5" fill="url(#fg-foundations)"/>
34<a href="#region-foundations" class="fg-map__region"><text x="175" y="68.5" text-anchor="middle" fill="#d9c7a1">FOUNDATIONS</text></a><ellipse cx="480" cy="150" rx="142.3" ry="113.8" fill="url(#fg-neural-nets)"/><a href="#region-neural-nets" class="fg-map__region"><text x="480" y="32.2" text-anchor="middle" fill="#ff6b35">NEURAL NETWORKS</text></a><ellipse cx="790" cy="175" rx="148.1" ry="118.5" fill="url(#fg-training)"/><a href="#region-training" class="fg-map__region"><text x="790" y="52.5" text-anchor="middle" fill="#ffc857">TRAINING</text></a><ellipse cx="1060" cy="230" rx="123" ry="98.4" fill="url(#fg-evaluation)"/><a href="#region-evaluation" class="fg-map__region"><text x="1060" y="127.6" text-anchor="middle" fill="#9ad97a">EVALUATION</text></a><ellipse cx="400" cy="470" rx="169.7" ry="135.8" fill="url(#fg-language)"/><a href="#region-language" class="fg-map__region"><text x="400" y="330.2" text-anchor="middle" fill="#3de0c5">LANGUAGE & LLMS</text></a><ellipse cx="730" cy="470" rx="130.2" ry="104.1" fill="url(#fg-vision)"/><a href="#region-vision" class="fg-map__region"><text x="730" y="361.9" text-anchor="middle" fill="#5aa9ff">VISION & MULTIMODAL</text></a><ellipse cx="140" cy="530" rx="111.1" ry="88.9" fill="url(#fg-agents)"/><a href="#region-agents" class="fg-map__region"><text x="140" y="437.1" text-anchor="middle" fill="#7b6cff">AGENTS & RL</text></a><ellipse cx="1020" cy="540" rx="142.3" ry="113.8" fill="url(#fg-shipping)"/><a href="#region-shipping" class="fg-map__region"><text x="1020" y="422.2" text-anchor="middle" fill="#ff5d8f">SHIPPING AI</text></a><g fill="none"><path d="M501.4 103.1 Q490.7 133.7 479.9 199.9" stroke="#ff6b35" class="fg-map__path"/><path d="M235.8 207.7 Q621.7 77.2 1007.7 224.7" stroke="#9ad97a" class="fg-map__path"/><path d="M426.1 412.8 Q268 398 109.9 501.3" stroke="#7b6cff" class="fg-map__path"/><path d="M93.9 536.5 Q101.9 511.9 109.9 501.3" stroke="#7b6cff" class="fg-map__path"/><path d="M324.5 462.4 Q217.2 442.6 109.9 501.3" stroke="#7b6cff" class="fg-map__path"/><path d="M156.1 494.8 Q140 516.1 123.9 565.2" stroke="#7b6cff" class="fg-map__path"/><path d="M454.5 427.5 Q431 458.7 407.5 530.6" stroke="#3de0c5" class="fg-map__path"/><path d="M501.4 103.1 Q454.5 238.1 407.5 530.6" stroke="#3de0c5" class="fg-map__path"/><path d="M501.4 103.1 Q626.3 110.7 751.2 217" stroke="#ffc857" class="fg-map__path"/><path d="M842 143.4 Q796.6 159.2 751.2 217" stroke="#ffc857" class="fg-map__path"/><path d="M197.2 141.4 Q216.5 160.8 235.8 207.7" stroke="#d9c7a1" class="fg-map__path"/><path d="M501.4 103.1 Q517.7 107 534.1 125.1" stroke="#ff6b35" class="fg-map__path"/><path d="M439.8 188.2 Q594.6 237.5 749.3 427.7" stroke="#5aa9ff" class="fg-map__path"/><path d="M373.9 527.2 Q349.2 480.1 324.5 462.4" stroke="#3de0c5" class="fg-map__path"/><path d="M475.5 477.6 Q400 442.7 324.5 462.4" stroke="#3de0c5" class="fg-map__path"/><path d="M501.4 103.1 Q470.6 126.8 439.8 188.2" stroke="#ff6b35" class="fg-map__path"/><path d="M501.4 103.1 Q588 216.2 674.5 477.8" stroke="#5aa9ff" class="fg-map__path"/><path d="M501.4 103.1 Q480.1 95.4 458.7 103.1" stroke="#ff6b35" class="fg-map__path"/><path d="M847.5 199.7 Q653.1 79.3 458.7 103.1" stroke="#ff6b35" class="fg-map__path"/><path d="M426 125 Q440.3 221.6 454.5 427.5" stroke="#3de0c5" class="fg-map__path"/><path d="M821.7 220.7 Q803.7 217.3 785.8 227.1" stroke="#ffc857" class="fg-map__path"/><path d="M821.7 220.7 Q607.1 230.7 392.5 409.4" stroke="#3de0c5" class="fg-map__path"/><path d="M116.4 211.8 Q132 163.9 147.7 143.1" stroke="#d9c7a1" class="fg-map__path"/><path d="M233.6 168.2 Q190.7 139.5 147.7 143.1" stroke="#d9c7a1" class="fg-map__path"/><path d="M855.1 170.9 Q848.6 151.7 842 143.4" stroke="#ffc857" class="fg-map__path"/><path d="M426.1 412.8 Q381.1 408.1 336.1 436.9" stroke="#3de0c5" class="fg-map__path"/><path d="M418.5 158.6 Q581 89.9 743.5 138.3" stroke="#ffc857" class="fg-map__path"/><path d="M812.4 125.9 Q777.9 119.5 743.5 138.3" stroke="#ffc857" class="fg-map__path"/><path d="M749.3 427.7 Q757.7 451.9 766.2 504.6" stroke="#5aa9ff" class="fg-map__path"/><path d="M235.8 207.7 Q501 246.7 766.2 504.6" stroke="#5aa9ff" class="fg-map__path"/><path d="M749.3 427.7 Q730 453.2 710.7 512.3" stroke="#5aa9ff" class="fg-map__path"/><path d="M812.4 125.9 Q926.9 231.6 1041.4 493.1" stroke="#ff5d8f" class="fg-map__path"/><path d="M472.2 450.7 Q468 467.3 463.9 503.1" stroke="#3de0c5" class="fg-map__path"/><path d="M439.6 522 Q451.8 507 463.9 503.1" stroke="#3de0c5" class="fg-map__path"/><path d="M842 143.4 Q785.5 146.2 729 193.6" stroke="#ffc857" class="fg-map__path"/><path d="M475.5 477.6 Q450.8 430.5 426.1 412.8" stroke="#3de0c5" class="fg-map__path"/><path d="M463.9 503.1 Q445 440.3 426.1 412.8" stroke="#3de0c5" class="fg-map__path"/><path d="M821.7 220.7 Q623.9 237.6 426.1 412.8" stroke="#3de0c5" class="fg-map__path"/><path d="M812.4 125.9 Q833.8 137.2 855.1 170.9" stroke="#ffc857" class="fg-map__path"/><path d="M116.4 211.8 Q308.9 85.4 501.4 103.1" stroke="#ff6b35" class="fg-map__path"/><path d="M749.3 427.7 Q767.4 435.9 785.5 462.2" stroke="#5aa9ff" class="fg-map__path"/><path d="M233.6 168.2 Q540.6 73.3 847.5 199.7" stroke="#ffc857" class="fg-map__path"/><path d="M501.4 103.1 Q460 112.9 418.5 158.6" stroke="#ff6b35" class="fg-map__path"/><path d="M123.9 565.2 Q155 530.9 186.1 523.5" stroke="#7b6cff" class="fg-map__path"/><path d="M812.4 125.9 Q817 156.2 821.7 220.7" stroke="#ffc857" class="fg-map__path"/><path d="M775.7 124 Q798.7 153.1 821.7 220.7" stroke="#ffc857" class="fg-map__path"/><path d="M426.1 412.8 Q377 428.7 327.8 489.3" stroke="#3de0c5" class="fg-map__path"/><path d="M426.1 412.8 Q393.2 403.5 360.4 418" stroke="#3de0c5" class="fg-map__path"/><path d="M454.5 427.5 Q407.5 405.7 360.4 418" stroke="#3de0c5" class="fg-map__path"/><path d="M197.2 141.4 Q175 170.8 152.8 238.6" stroke="#d9c7a1" class="fg-map__path"/><path d="M847.5 199.7 Q786.8 159.2 726.1 164.3" stroke="#ffc857" class="fg-map__path"/><path d="M116.4 211.8 Q136.2 301.9 156.1 494.8" stroke="#7b6cff" class="fg-map__path"/><path d="M439.8 188.2 Q432.9 145 426 125" stroke="#ff6b35" class="fg-map__path"/><path d="M202.3 236.9 Q314.1 135.9 426 125" stroke="#ff6b35" class="fg-map__path"/><path d="M156.1 494.8 Q163.1 515 170.1 558.7" stroke="#7b6cff" class="fg-map__path"/><path d="M501.4 103.1 Q510.7 130 520.1 188.3" stroke="#ff6b35" class="fg-map__path"/><path d="M114.2 172.3 Q445 28.8 775.7 124" stroke="#ffc857" class="fg-map__path"/><path d="M749.3 427.7 Q721.6 421.5 693.8 435.4" stroke="#5aa9ff" class="fg-map__path"/><path d="M235.8 207.7 Q464.8 229.5 693.8 435.4" stroke="#5aa9ff" class="fg-map__path"/><path d="M407.5 530.6 Q376.5 509.9 345.5 512.5" stroke="#3de0c5" class="fg-map__path"/><path d="M479.9 199.9 Q510.7 166 541.4 158.7" stroke="#ff6b35" class="fg-map__path"/><path d="M116.4 211.8 Q156.8 157.3 197.2 141.4" stroke="#d9c7a1" class="fg-map__path"/><path d="M373.9 527.2 Q406.8 512.7 439.6 522" stroke="#3de0c5" class="fg-map__path"/><path d="M426.1 412.8 Q260 410.8 93.9 536.5" stroke="#7b6cff" class="fg-map__path"/><path d="M407.5 530.6 Q441.5 488.6 475.5 477.6" stroke="#3de0c5" class="fg-map__path"/><path d="M454.5 427.5 Q465 442.8 475.5 477.6" stroke="#3de0c5" class="fg-map__path"/><path d="M116.4 211.8 Q115.3 184.9 114.2 172.3" stroke="#d9c7a1" class="fg-map__path"/></g><circle cx="1012.6" cy="208" r="2.4" fill="#9ad97a" fill-opacity="0.55"/><circle cx="782.9" cy="460.7" r="1.6" fill="#5aa9ff" fill-opacity="0.3"/><circle cx="146.8" cy="244" r="1.6" fill="#d9c7a1" fill-opacity="0.3"/><circle cx="159.7" cy="550.6" r="2.4" fill="#7b6cff" fill-opacity="0.55"/><circle cx="721.7" cy="157.6" r="1.6" fill="#ffc857" fill-opacity="0.3"/><circle cx="775.3" cy="185.7" r="2.4" fill="#ffc857" fill-opacity="0.55"/><circle cx="856.3" cy="150.5" r="2.4" fill="#ffc857" fill-opacity="0.55"/><circle cx="1068.7" cy="531.4" r="2.4" fill="#ff5d8f" fill-opacity="0.55"/><circle cx="1057.6" cy="551" r="1.6" fill="#ff5d8f" fill-opacity="0.3"/><circle cx="990.3" cy="573.8" r="2.4" fill="#ff5d8f" fill-opacity="0.55"/><circle cx="981" cy="552.9" r="2.4" fill="#ff5d8f" fill-opacity="0.55"/><circle cx="1082.9" cy="516.8" r="2.4" fill="#ff5d8f" fill-opacity="0.55"/><circle cx="1038.7" cy="508" r="2.4" fill="#ff5d8f" fill-opacity="0.55"/><circle cx="1010.8" cy="567.2" r="2.4" fill="#ff5d8f" fill-opacity="0.55"/><circle cx="955.2" cy="523.5" r="2.4" fill="#ff5d8f" fill-opacity="0.55"/><circle cx="992.8" cy="592.1" r="1.6" fill="#ff5d8f" fill-opacity="0.3"/><circle cx="316.9" cy="463.8" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="156.8" cy="572.9" r="1.6" fill="#7b6cff" fill-opacity="0.3"/><circle cx="149.8" cy="134" r="1.6" fill="#d9c7a1" fill-opacity="0.3"/><circle cx="759.9" cy="454.8" r="1.6" fill="#5aa9ff" fill-opacity="0.3"/><circle cx="467.3" cy="429.1" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="116.4" cy="191.9" r="2.4" fill="#d9c7a1" fill-opacity="0.55"/><circle cx="384.7" cy="537.4" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="442.1" cy="498.4" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="353.9" cy="411.3" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="484.5" cy="488.5" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="1115.5" cy="220.2" r="1.6" fill="#9ad97a" fill-opacity="0.3"/><circle cx="1026.1" cy="268.5" r="2.4" fill="#9ad97a" fill-opacity="0.55"/><circle cx="732.2" cy="450" r="2.4" fill="#5aa9ff" fill-opacity="0.55"/><circle cx="321.1" cy="502.4" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="509.2" cy="135.1" r="2.4" fill="#ff6b35" fill-opacity="0.55"/><circle cx="697.7" cy="506.7" r="1.6" fill="#5aa9ff" fill-opacity="0.3"/><circle cx="413.4" cy="470" r="2.4" fill="#3de0c5" fill-opacity="0.55"/>
34<circle cx="1078.5" cy="249.3" r="2.4" fill="#9ad97a" fill-opacity="0.55"/><circle cx="761.3" cy="229.9" r="1.6" fill="#ffc857" fill-opacity="0.3"/><circle cx="995.7" cy="486" r="1.6" fill="#ff5d8f" fill-opacity="0.3"/><circle cx="495.2" cy="95.9" r="2.4" fill="#ff6b35" fill-opacity="0.55"/><circle cx="242" cy="218.2" r="2.4" fill="#d9c7a1" fill-opacity="0.55"/><circle cx="431.2" cy="402.7" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="1052.2" cy="181.8" r="2.4" fill="#9ad97a" fill-opacity="0.55"/><circle cx="435.9" cy="129.5" r="2.4" fill="#ff6b35" fill-opacity="0.55"/><circle cx="1013.1" cy="497.7" r="1.6" fill="#ff5d8f" fill-opacity="0.3"/><circle cx="792.1" cy="155.6" r="2.4" fill="#ffc857" fill-opacity="0.55"/><circle cx="764.4" cy="118" r="1.6" fill="#ffc857" fill-opacity="0.3"/><circle cx="89.4" cy="506.5" r="1.6" fill="#7b6cff" fill-opacity="0.3"/><circle cx="160.6" cy="200.6" r="2.4" fill="#d9c7a1" fill-opacity="0.55"/><circle cx="199.3" cy="519.6" r="2.4" fill="#7b6cff" fill-opacity="0.55"/><circle cx="1107.9" cy="262.3" r="1.6" fill="#9ad97a" fill-opacity="0.3"/><circle cx="343.3" cy="471.9" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="434.2" cy="537.2" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="515.2" cy="199.1" r="1.6" fill="#ff6b35" fill-opacity="0.3"/><circle cx="100.6" cy="205.7" r="2.4" fill="#d9c7a1" fill-opacity="0.55"/><circle cx="1062" cy="568.3" r="2.4" fill="#ff5d8f" fill-opacity="0.55"/><circle cx="1026.5" cy="225.3" r="2.4" fill="#9ad97a" fill-opacity="0.55"/><circle cx="441.3" cy="437.1" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="963.5" cy="541.9" r="2.4" fill="#ff5d8f" fill-opacity="0.55"/><circle cx="995.6" cy="232.1" r="2.4" fill="#9ad97a" fill-opacity="0.55"/><circle cx="1084.6" cy="567.2" r="1.6" fill="#ff5d8f" fill-opacity="0.3"/><circle cx="531.7" cy="140.9" r="1.6" fill="#ff6b35" fill-opacity="0.3"/><circle cx="317.2" cy="438.6" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="411.3" cy="132.5" r="1.6" fill="#ff6b35" fill-opacity="0.3"/><circle cx="744.5" cy="470" r="2.4" fill="#5aa9ff" fill-opacity="0.55"/><circle cx="488.5" cy="448.2" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="1107" cy="192.6" r="2.4" fill="#9ad97a" fill-opacity="0.55"/><circle cx="948.2" cy="555.1" r="1.6" fill="#ff5d8f" fill-opacity="0.3"/><circle cx="546.7" cy="125.3" r="1.6" fill="#ff6b35" fill-opacity="0.3"/><circle cx="758.7" cy="210.6" r="2.4" fill="#ffc857" fill-opacity="0.55"/><circle cx="352.7" cy="534.6" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="217.3" cy="137.4" r="1.6" fill="#d9c7a1" fill-opacity="0.3"/><circle cx="380.2" cy="396.1" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="722.6" cy="424" r="1.6" fill="#5aa9ff" fill-opacity="0.3"/><circle cx="1060.8" cy="489.3" r="1.6" fill="#ff5d8f" fill-opacity="0.3"/><circle cx="304.3" cy="479.5" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="477.8" cy="514.2" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="1033" cy="600.4" r="1.6" fill="#ff5d8f" fill-opacity="0.3"/><circle cx="463.1" cy="410.7" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="217.7" cy="156" r="2.4" fill="#d9c7a1" fill-opacity="0.55"/><circle cx="775.7" cy="500.8" r="2.4" fill="#5aa9ff" fill-opacity="0.55"/><circle cx="1056.9" cy="284.3" r="1.6" fill="#9ad97a" fill-opacity="0.3"/><circle cx="668.6" cy="472" r="1.6" fill="#5aa9ff" fill-opacity="0.3"/><circle cx="403.6" cy="548.5" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="188.4" cy="252.6" r="1.6" fill="#d9c7a1" fill-opacity="0.3"/><circle cx="1033.8" cy="575.2" r="2.4" fill="#ff5d8f" fill-opacity="0.55"/><circle cx="958.6" cy="501.9" r="2.4" fill="#ff5d8f" fill-opacity="0.55"/><circle cx="111.3" cy="150.5" r="1.6" fill="#d9c7a1" fill-opacity="0.3"/><circle cx="256.5" cy="184.6" r="1.6" fill="#d9c7a1" fill-opacity="0.3"/><circle cx="1075.2" cy="230" r="2.4" fill="#9ad97a" fill-opacity="0.55"/><circle cx="330.4" cy="413.5" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="1098.6" cy="534.8" r="1.6" fill="#ff5d8f" fill-opacity="0.3"/><circle cx="499.9" cy="474.1" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="858.1" cy="203.6" r="1.6" fill="#ffc857" fill-opacity="0.3"/><circle cx="714.4" cy="191" r="2.4" fill="#ffc857" fill-opacity="0.55"/><circle cx="782.8" cy="130.4" r="2.4" fill="#ffc857" fill-opacity="0.55"/><circle cx="322.2" cy="521.4" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="494.2" cy="150" r="2.4" fill="#ff6b35" fill-opacity="0.55"/><circle cx="833" cy="121.5" r="1.6" fill="#ffc857" fill-opacity="0.3"/><circle cx="414.1" cy="389.3" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="397.2" cy="517.8" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="172.1" cy="239.4" r="1.6" fill="#d9c7a1" fill-opacity="0.3"/><circle cx="458.2" cy="537.7" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="807.1" cy="192.8" r="2.4" fill="#ffc857" fill-opacity="0.55"/><circle cx="803.7" cy="238.6" r="1.6" fill="#ffc857" fill-opacity="0.3"/><circle cx="1015.4" cy="187.2" r="2.4" fill="#9ad97a" fill-opacity="0.55"/><circle cx="725.2" cy="134.9" r="1.6" fill="#ffc857" fill-opacity="0.3"/><circle cx="872.8" cy="169.5" r="1.6" fill="#ffc857" fill-opacity="0.3"/><circle cx="134.4" cy="151.1" r="2.4" fill="#d9c7a1" fill-opacity="0.55"/><circle cx="732.7" cy="224.5" r="1.6" fill="#ffc857" fill-opacity="0.3"/>
34<circle cx="790.4" cy="106.6" r="1.6" fill="#ffc857" fill-opacity="0.3"/><circle cx="118.7" cy="238.7" r="2.4" fill="#d9c7a1" fill-opacity="0.55"/><circle cx="759.1" cy="170.6" r="2.4" fill="#ffc857" fill-opacity="0.55"/><circle cx="175.4" cy="122.8" r="2.4" fill="#d9c7a1" fill-opacity="0.55"/><circle cx="1061.2" cy="507.2" r="2.4" fill="#ff5d8f" fill-opacity="0.55"/><circle cx="819" cy="160.2" r="2.4" fill="#ffc857" fill-opacity="0.55"/><circle cx="1130.7" cy="237.6" r="2.4" fill="#9ad97a" fill-opacity="0.55"/><circle cx="965.7" cy="587" r="2.4" fill="#ff5d8f" fill-opacity="0.55"/><circle cx="213.9" cy="201.4" r="2.4" fill="#d9c7a1" fill-opacity="0.55"/><circle cx="980.9" cy="502.5" r="2.4" fill="#ff5d8f" fill-opacity="0.55"/><circle cx="1017.2" cy="587.6" r="2.4" fill="#ff5d8f" fill-opacity="0.55"/><circle cx="188.8" cy="190" r="2.4" fill="#d9c7a1" fill-opacity="0.55"/><circle cx="451.1" cy="205.3" r="1.6" fill="#ff6b35" fill-opacity="0.3"/><circle cx="360.7" cy="432.4" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="191.8" cy="207.5" r="2.4" fill="#d9c7a1" fill-opacity="0.55"/><circle cx="462.1" cy="476.7" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="1082" cy="546.7" r="2.4" fill="#ff5d8f" fill-opacity="0.55"/><circle cx="720.1" cy="499.6" r="1.6" fill="#5aa9ff" fill-opacity="0.3"/><circle cx="386.1" cy="480.2" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="454.2" cy="92.6" r="1.6" fill="#ff6b35" fill-opacity="0.3"/><circle cx="299.1" cy="451.4" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="1000.2" cy="263.3" r="1.6" fill="#9ad97a" fill-opacity="0.3"/><circle cx="1020.4" cy="475.1" r="1.6" fill="#ff5d8f" fill-opacity="0.3"/><circle cx="144.6" cy="185.7" r="2.4" fill="#d9c7a1" fill-opacity="0.55"/><circle cx="548.6" cy="178.8" r="1.6" fill="#ff6b35" fill-opacity="0.3"/><circle cx="774.8" cy="434.4" r="1.6" fill="#5aa9ff" fill-opacity="0.3"/><circle cx="848.2" cy="226.3" r="1.6" fill="#ffc857" fill-opacity="0.3"/><circle cx="834.3" cy="204.9" r="2.4" fill="#ffc857" fill-opacity="0.55"/><circle cx="1076.3" cy="171.9" r="2.4" fill="#9ad97a" fill-opacity="0.55"/><circle cx="103.9" cy="571.1" r="1.6" fill="#7b6cff" fill-opacity="0.3"/><circle cx="131.7" cy="478.5" r="2.4" fill="#7b6cff" fill-opacity="0.55"/><circle cx="232.2" cy="240.5" r="1.6" fill="#d9c7a1" fill-opacity="0.3"/><circle cx="1075.2" cy="588.7" r="2.4" fill="#ff5d8f" fill-opacity="0.55"/><circle cx="727" cy="521.8" r="1.6" fill="#5aa9ff" fill-opacity="0.3"/><circle cx="89.1" cy="183.6" r="1.6" fill="#d9c7a1" fill-opacity="0.3"/><circle cx="702.6" cy="168.5" r="2.4" fill="#ffc857" fill-opacity="0.55"/><circle cx="403.8" cy="166.1" r="1.6" fill="#ff6b35" fill-opacity="0.3"/><circle cx="244.6" cy="147.7" r="1.6" fill="#d9c7a1" fill-opacity="0.3"/><circle cx="490.8" cy="428.8" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="347.4" cy="499.3" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="448.8" cy="145.6" r="2.4" fill="#ff6b35" fill-opacity="0.55"/><circle cx="416.3" cy="487" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="367.5" cy="550.1" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="134.6" cy="203.3" r="2.4" fill="#d9c7a1" fill-opacity="0.55"/><circle cx="203.6" cy="175.5" r="2.4" fill="#d9c7a1" fill-opacity="0.55"/><circle cx="123.1" cy="542.4" r="2.4" fill="#7b6cff" fill-opacity="0.55"/><circle cx="356" cy="392.8" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="860.8" cy="132" r="2.4" fill="#ffc857" fill-opacity="0.55"/><circle cx="730.4" cy="177" r="1.6" fill="#ffc857" fill-opacity="0.3"/><circle cx="523.3" cy="96.1" r="1.6" fill="#ff6b35" fill-opacity="0.3"/><circle cx="937.1" cy="533.9" r="2.4" fill="#ff5d8f" fill-opacity="0.55"/><circle cx="711.1" cy="440.9" r="1.6" fill="#5aa9ff" fill-opacity="0.3"/><circle cx="687.5" cy="429.2" r="1.6" fill="#5aa9ff" fill-opacity="0.3"/><circle cx="1091.4" cy="214" r="2.4" fill="#9ad97a" fill-opacity="0.55"/><circle cx="104.3" cy="524.9" r="2.4" fill="#7b6cff" fill-opacity="0.55"/><circle cx="1004.7" cy="607.2" r="2.4" fill="#ff5d8f" fill-opacity="0.55"/><circle cx="1087.2" cy="499.2" r="2.4" fill="#ff5d8f" fill-opacity="0.55"/><circle cx="1097.8" cy="282.8" r="2.4" fill="#9ad97a" fill-opacity="0.55"/><circle cx="833.4" cy="140.4" r="1.6" fill="#ffc857" fill-opacity="0.3"/>
34<circle cx="1049.6" cy="261.1" r="2.4" fill="#9ad97a" fill-opacity="0.55"/><circle cx="194.4" cy="156.8" r="2.4" fill="#d9c7a1" fill-opacity="0.55"/><circle cx="448.5" cy="185.9" r="2.4" fill="#ff6b35" fill-opacity="0.55"/><circle cx="127.3" cy="129.4" r="2.4" fill="#d9c7a1" fill-opacity="0.55"/><circle cx="974" cy="481.5" r="1.6" fill="#ff5d8f" fill-opacity="0.3"/><circle cx="159.2" cy="259.6" r="1.6" fill="#d9c7a1" fill-opacity="0.3"/><circle cx="770.8" cy="481.9" r="1.6" fill="#5aa9ff" fill-opacity="0.3"/><circle cx="1104.3" cy="558.5" r="2.4" fill="#ff5d8f" fill-opacity="0.55"/><circle cx="472.7" cy="105.1" r="2.4" fill="#ff6b35" fill-opacity="0.55"/><circle cx="414.4" cy="418.9" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="493.8" cy="214.1" r="1.6" fill="#ff6b35" fill-opacity="0.3"/><circle cx="498.5" cy="503.4" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="941.3" cy="572.3" r="1.6" fill="#ff5d8f" fill-opacity="0.3"/><circle cx="787.1" cy="225.2" r="2.4" fill="#ffc857" fill-opacity="0.55"/><circle cx="524.6" cy="180.1" r="2.4" fill="#ff6b35" fill-opacity="0.55"/><circle cx="687.7" cy="484" r="1.6" fill="#5aa9ff" fill-opacity="0.3"/><circle cx="262.4" cy="209.2" r="1.6" fill="#d9c7a1" fill-opacity="0.3"/><circle cx="191.1" cy="564.5" r="2.4" fill="#7b6cff" fill-opacity="0.55"/><circle cx="714.9" cy="481.1" r="2.4" fill="#5aa9ff" fill-opacity="0.55"/><circle cx="414.8" cy="109.6" r="1.6" fill="#ff6b35" fill-opacity="0.3"/><circle cx="563.4" cy="144.5" r="2.4" fill="#ff6b35" fill-opacity="0.55"/><circle cx="93.5" cy="223.5" r="1.6" fill="#d9c7a1" fill-opacity="0.3"/><circle cx="797.3" cy="477.2" r="1.6" fill="#5aa9ff" fill-opacity="0.3"/><circle cx="470.3" cy="178.9" r="2.4" fill="#ff6b35" fill-opacity="0.55"/><circle cx="207.2" cy="120.5" r="1.6" fill="#d9c7a1" fill-opacity="0.3"/><circle cx="1022" cy="521.6" r="2.4" fill="#ff5d8f" fill-opacity="0.55"/><circle cx="773.9" cy="245.8" r="1.6" fill="#ffc857" fill-opacity="0.3"/><circle cx="210.3" cy="259.4" r="2.4" fill="#d9c7a1" fill-opacity="0.55"/><circle cx="298.4" cy="498.8" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="986.2" cy="211.2" r="1.6" fill="#9ad97a" fill-opacity="0.3"/><circle cx="741.5" cy="113.3" r="1.6" fill="#ffc857" fill-opacity="0.3"/><circle cx="878.9" cy="194.5" r="1.6" fill="#ffc857" fill-opacity="0.3"/><circle cx="89.4" cy="157.5" r="1.6" fill="#d9c7a1" fill-opacity="0.3"/><circle cx="433.2" cy="516.4" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="422.3" cy="199.9" r="1.6" fill="#ff6b35" fill-opacity="0.3"/><circle cx="480.4" cy="81.1" r="2.4" fill="#ff6b35" fill-opacity="0.55"/><circle cx="1051.1" cy="472.9" r="1.6" fill="#ff5d8f" fill-opacity="0.3"/><circle cx="451" cy="393.3" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="538.6" cy="201.7" r="1.6" fill="#ff6b35" fill-opacity="0.3"/><circle cx="461.6" cy="121.6" r="2.4" fill="#ff6b35" fill-opacity="0.55"/><circle cx="71.2" cy="532.3" r="1.6" fill="#7b6cff" fill-opacity="0.3"/><circle cx="707.1" cy="209" r="2.4" fill="#ffc857" fill-opacity="0.55"/><circle cx="266.5" cy="167.4" r="1.6" fill="#d9c7a1" fill-opacity="0.3"/><circle cx="1054.1" cy="607" r="1.6" fill="#ff5d8f" fill-opacity="0.3"/><circle cx="822.8" cy="104.3" r="2.4" fill="#ffc857" fill-opacity="0.55"/><circle cx="748.9" cy="188.6" r="2.4" fill="#ffc857" fill-opacity="0.55"/><circle cx="392" cy="143.5" r="1.6" fill="#ff6b35" fill-opacity="0.3"/><circle cx="747.6" cy="488.4" r="2.4" fill="#5aa9ff" fill-opacity="0.55"/><circle cx="673" cy="501.7" r="1.6" fill="#5aa9ff" fill-opacity="0.3"/><circle cx="745.6" cy="414.6" r="1.6" fill="#5aa9ff" fill-opacity="0.3"/><circle cx="698.1" cy="465.5" r="2.4" fill="#5aa9ff" fill-opacity="0.55"/><circle cx="239.2" cy="196.9" r="1.6" fill="#d9c7a1" fill-opacity="0.3"/><circle cx="173.5" cy="512.9" r="2.4" fill="#7b6cff" fill-opacity="0.55"/><circle cx="335.1" cy="453.4" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="766" cy="520.3" r="2.4" fill="#5aa9ff" fill-opacity="0.55"/><circle cx="659.6" cy="452" r="1.6" fill="#5aa9ff" fill-opacity="0.3"/><circle cx="937.4" cy="508.7" r="2.4" fill="#ff5d8f" fill-opacity="0.55"/><circle cx="126.1" cy="256.8" r="1.6" fill="#d9c7a1" fill-opacity="0.3"/><circle cx="154.5" cy="113.6" r="1.6" fill="#d9c7a1" fill-opacity="0.3"/><circle cx="427.4" cy="554.8" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="307.6" cy="421.9" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="255.4" cy="235.7" r="2.4" fill="#d9c7a1" fill-opacity="0.55"/><circle cx="509.5" cy="455.4" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="331.1" cy="540.5" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="360.9" cy="482.9" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="750.4" cy="435.2" r="2.4" fill="#5aa9ff" fill-opacity="0.55"/><circle cx="990.7" cy="535.8" r="2.4" fill="#ff5d8f" fill-opacity="0.55"/><circle cx="190.1" cy="490.1" r="1.6" fill="#7b6cff" fill-opacity="0.3"/><circle cx="76.1" cy="199.9" r="1.6" fill="#d9c7a1" fill-opacity="0.3"/><circle cx="240.2" cy="128.7" r="2.4" fill="#d9c7a1" fill-opacity="0.55"/><circle cx="551.3" cy="106.7" r="2.4" fill="#ff6b35" fill-opacity="0.55"/><circle cx="798.3" cy="444.7" r="2.4" fill="#5aa9ff" fill-opacity="0.55"/><circle cx="177.1" cy="170.9" r="2.4" fill="#d9c7a1" fill-opacity="0.55"/><circle cx="829.6" cy="186.6" r="2.4" fill="#ffc857" fill-opacity="0.55"/><circle cx="178.7" cy="271.2" r="1.6" fill="#d9c7a1" fill-opacity="0.3"/>
34<circle cx="825.9" cy="245.6" r="1.6" fill="#ffc857" fill-opacity="0.3"/><circle cx="703" cy="142" r="2.4" fill="#ffc857" fill-opacity="0.55"/><circle cx="883" cy="152.1" r="1.6" fill="#ffc857" fill-opacity="0.3"/><circle cx="189.3" cy="226.5" r="2.4" fill="#d9c7a1" fill-opacity="0.55"/><circle cx="740.3" cy="242.9" r="1.6" fill="#ffc857" fill-opacity="0.3"/><circle cx="463.1" cy="446.7" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="700.4" cy="526.6" r="2.4" fill="#5aa9ff" fill-opacity="0.55"/><circle cx="519.9" cy="161.6" r="2.4" fill="#ff6b35" fill-opacity="0.55"/><circle cx="165.5" cy="218.2" r="1.6" fill="#d9c7a1" fill-opacity="0.3"/><circle cx="497.2" cy="168" r="2.4" fill="#ff6b35" fill-opacity="0.55"/><circle cx="391.2" cy="380.2" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="420" cy="152" r="2.4" fill="#ff6b35" fill-opacity="0.55"/><circle cx="780.3" cy="203.7" r="2.4" fill="#ffc857" fill-opacity="0.55"/><circle cx="482.9" cy="531.9" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="285.8" cy="469.1" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="1131.7" cy="203.5" r="2.4" fill="#9ad97a" fill-opacity="0.55"/><circle cx="103.1" cy="131.6" r="2.4" fill="#d9c7a1" fill-opacity="0.55"/><circle cx="1028.9" cy="289.4" r="2.4" fill="#9ad97a" fill-opacity="0.55"/><circle cx="769.1" cy="97.3" r="1.6" fill="#ffc857" fill-opacity="0.3"/><circle cx="871.8" cy="221.5" r="2.4" fill="#ffc857" fill-opacity="0.55"/><circle cx="431.2" cy="87.9" r="2.4" fill="#ff6b35" fill-opacity="0.55"/><circle cx="463.8" cy="221.3" r="1.6" fill="#ff6b35" fill-opacity="0.3"/><circle cx="278.3" cy="194.2" r="1.6" fill="#d9c7a1" fill-opacity="0.3"/><circle cx="94.6" cy="243.2" r="1.6" fill="#d9c7a1" fill-opacity="0.3"/><circle cx="703.6" cy="411.3" r="1.6" fill="#5aa9ff" fill-opacity="0.3"/><circle cx="1040.2" cy="199.5" r="1.6" fill="#9ad97a" fill-opacity="0.3"/><circle cx="372.7" cy="522.2" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="375.6" cy="415.8" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="418.8" cy="437.9" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="569.5" cy="169.6" r="1.6" fill="#ff6b35" fill-opacity="0.3"/><circle cx="1032.3" cy="168.4" r="1.6" fill="#9ad97a" fill-opacity="0.3"/><circle cx="120.6" cy="220.3" r="1.6" fill="#d9c7a1" fill-opacity="0.3"/><circle cx="128.9" cy="563.2" r="1.6" fill="#7b6cff" fill-opacity="0.3"/><circle cx="118.9" cy="497.4" r="2.4" fill="#7b6cff" fill-opacity="0.55"/><circle cx="1108.2" cy="518.2" r="1.6" fill="#ff5d8f" fill-opacity="0.3"/><circle cx="1133.7" cy="261" r="2.4" fill="#9ad97a" fill-opacity="0.55"/><circle cx="800.2" cy="499.5" r="1.6" fill="#5aa9ff" fill-opacity="0.3"/><circle cx="396.5" cy="184.3" r="1.6" fill="#ff6b35" fill-opacity="0.3"/><circle cx="748.7" cy="135.4" r="1.6" fill="#ffc857" fill-opacity="0.3"/><circle cx="136.6" cy="588" r="1.6" fill="#7b6cff" fill-opacity="0.3"/><circle cx="689.5" cy="185" r="2.4" fill="#ffc857" fill-opacity="0.55"/><circle cx="793.8" cy="257.5" r="2.4" fill="#ffc857" fill-opacity="0.55"/><circle cx="856.3" cy="112.7" r="2.4" fill="#ffc857" fill-opacity="0.55"/><circle cx="513" cy="78.8" r="1.6" fill="#ff6b35" fill-opacity="0.3"/><circle cx="716.9" cy="115.6" r="1.6" fill="#ffc857" fill-opacity="0.3"/><circle cx="1047.5" cy="526" r="2.4" fill="#ff5d8f" fill-opacity="0.55"/><circle cx="131.9" cy="170" r="2.4" fill="#d9c7a1" fill-opacity="0.55"/><circle cx="1036.2" cy="556.9" r="2.4" fill="#ff5d8f" fill-opacity="0.55"/><circle cx="516.2" cy="221.1" r="2.4" fill="#ff6b35" fill-opacity="0.55"/><circle cx="92.4" cy="484.3" r="2.4" fill="#7b6cff" fill-opacity="0.55"/><circle cx="972.8" cy="604.4" r="1.6" fill="#ff5d8f" fill-opacity="0.3"/><circle cx="978.4" cy="520.7" r="1.6" fill="#ff5d8f" fill-opacity="0.3"/><circle cx="1000.2" cy="466.3" r="1.6" fill="#ff5d8f" fill-opacity="0.3"/><circle cx="967.5" cy="569.2" r="2.4" fill="#ff5d8f" fill-opacity="0.55"/><circle cx="1034.3" cy="489" r="2.4" fill="#ff5d8f" fill-opacity="0.55"/><circle cx="1097.6" cy="584.1" r="1.6" fill="#ff5d8f" fill-opacity="0.3"/><circle cx="215.4" cy="538.1" r="2.4" fill="#7b6cff" fill-opacity="0.55"/><circle cx="924.6" cy="549.5" r="1.6" fill="#ff5d8f" fill-opacity="0.3"/><circle cx="1082.9" cy="480.9" r="1.6" fill="#ff5d8f" fill-opacity="0.3"/><circle cx="1023.6" cy="618.3" r="2.4" fill="#ff5d8f" fill-opacity="0.55"/><circle cx="809.8" cy="141.2" r="2.4" fill="#ffc857" fill-opacity="0.55"/><circle cx="1119.6" cy="544.1" r="2.4" fill="#ff5d8f" fill-opacity="0.55"/><circle cx="942.4" cy="591.3" r="2.4" fill="#ff5d8f" fill-opacity="0.55"/><circle cx="1033.3" cy="540" r="2.4" fill="#ff5d8f" fill-opacity="0.55"/><circle cx="950.6" cy="483.7" r="2.4" fill="#ff5d8f" fill-opacity="0.55"/><circle cx="1053.1" cy="586.3" r="2.4" fill="#ff5d8f" fill-opacity="0.55"/><circle cx="1034" cy="459.6" r="1.6" fill="#ff5d8f" fill-opacity="0.3"/><circle cx="392.3" cy="116.8" r="2.4" fill="#ff6b35" fill-opacity="0.55"/><circle cx="804.6" cy="212.2" r="2.4" fill="#ffc857" fill-opacity="0.55"/><circle cx="855.3" cy="182" r="2.4" fill="#ffc857" fill-opacity="0.55"/><circle cx="76.2" cy="565.5" r="1.6" fill="#7b6cff" fill-opacity="0.3"/><circle cx="157.4" cy="468" r="2.4" fill="#7b6cff" fill-opacity="0.55"/><circle cx="464.8" cy="497.3" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="895" cy="179.3" r="1.6" fill="#ffc857" fill-opacity="0.3"/><circle cx="652" cy="486.4" r="2.4" fill="#5aa9ff" fill-opacity="0.55"/><circle cx="189.9" cy="137.2" r="1.6" fill="#d9c7a1" fill-opacity="0.3"/><circle cx="413.9" cy="505.3" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="708.2" cy="229.1" r="1.6" fill="#ffc857" fill-opacity="0.3"/><circle cx="402" cy="451.5" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="804.8" cy="90.2" r="1.6" fill="#ffc857" fill-opacity="0.3"/><circle cx="573.7" cy="126.9" r="1.6" fill="#ff6b35" fill-opacity="0.3"/>
34<circle cx="1078.1" cy="607.5" r="1.6" fill="#ff5d8f" fill-opacity="0.3"/><circle cx="328" cy="485.2" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="774.3" cy="414.9" r="1.6" fill="#5aa9ff" fill-opacity="0.3"/><circle cx="485.6" cy="408.5" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="189.5" cy="106.6" r="1.6" fill="#d9c7a1" fill-opacity="0.3"/><circle cx="429.9" cy="218.4" r="1.6" fill="#ff6b35" fill-opacity="0.3"/><circle cx="157" cy="162.3" r="2.4" fill="#d9c7a1" fill-opacity="0.55"/><circle cx="235.2" cy="259.9" r="1.6" fill="#d9c7a1" fill-opacity="0.3"/><circle cx="744.1" cy="535.6" r="1.6" fill="#5aa9ff" fill-opacity="0.3"/><circle cx="180.3" cy="586.3" r="1.6" fill="#7b6cff" fill-opacity="0.3"/><circle cx="851.2" cy="246.1" r="1.6" fill="#ffc857" fill-opacity="0.3"/><circle cx="70.7" cy="170.8" r="1.6" fill="#d9c7a1" fill-opacity="0.3"/><circle cx="978.2" cy="247.3" r="2.4" fill="#9ad97a" fill-opacity="0.55"/><circle cx="663.2" cy="428.6" r="1.6" fill="#5aa9ff" fill-opacity="0.3"/><circle cx="771.7" cy="146.8" r="2.4" fill="#ffc857" fill-opacity="0.55"/><circle cx="459" cy="71.7" r="1.6" fill="#ff6b35" fill-opacity="0.3"/><circle cx="440.9" cy="419.1" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="388.8" cy="562.2" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="268.9" cy="147.4" r="2.4" fill="#d9c7a1" fill-opacity="0.55"/><circle cx="225.5" cy="181.1" r="1.6" fill="#d9c7a1" fill-opacity="0.3"/><circle cx="746.2" cy="154.7" r="2.4" fill="#ffc857" fill-opacity="0.55"/><circle cx="523.7" cy="115.2" r="2.4" fill="#ff6b35" fill-opacity="0.55"/><circle cx="919.4" cy="521.5" r="2.4" fill="#ff5d8f" fill-opacity="0.55"/><circle cx="1106.5" cy="172.1" r="2.4" fill="#9ad97a" fill-opacity="0.55"/><circle cx="141.4" cy="272.8" r="1.6" fill="#d9c7a1" fill-opacity="0.3"/><circle cx="683.9" cy="155.5" r="1.6" fill="#ffc857" fill-opacity="0.3"/><circle cx="142.5" cy="507.6" r="2.4" fill="#7b6cff" fill-opacity="0.55"/><circle cx="562.4" cy="196.8" r="2.4" fill="#ff6b35" fill-opacity="0.55"/><circle cx="358.3" cy="450.7" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="815.4" cy="464.3" r="1.6" fill="#5aa9ff" fill-opacity="0.3"/><circle cx="330" cy="395.5" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="209.3" cy="238" r="1.6" fill="#d9c7a1" fill-opacity="0.3"/><circle cx="61.2" cy="509.9" r="2.4" fill="#7b6cff" fill-opacity="0.55"/><circle cx="1102.8" cy="242.5" r="2.4" fill="#9ad97a" fill-opacity="0.55"/><circle cx="1110.6" cy="498.9" r="2.4" fill="#ff5d8f" fill-opacity="0.55"/><circle cx="378.7" cy="160.1" r="1.6" fill="#ff6b35" fill-opacity="0.3"/><circle cx="144.2" cy="225" r="1.6" fill="#d9c7a1" fill-opacity="0.3"/><circle cx="1006.1" cy="550.2" r="2.4" fill="#ff5d8f" fill-opacity="0.55"/><circle cx="515.3" cy="487.2" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="107.9" cy="172.9" r="1.6" fill="#d9c7a1" fill-opacity="0.3"/><circle cx="987.6" cy="619.9" r="2.4" fill="#ff5d8f" fill-opacity="0.55"/><circle cx="299.8" cy="519.9" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="976.1" cy="463" r="2.4" fill="#ff5d8f" fill-opacity="0.55"/><circle cx="185.7" cy="543.4" r="2.4" fill="#7b6cff" fill-opacity="0.55"/><circle cx="413" cy="530.5" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="745" cy="508.3" r="1.6" fill="#5aa9ff" fill-opacity="0.3"/><circle cx="448.9" cy="461.4" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="1074.8" cy="298.8" r="1.6" fill="#9ad97a" fill-opacity="0.3"/><circle cx="129.5" cy="110.2" r="1.6" fill="#d9c7a1" fill-opacity="0.3"/><circle cx="671" cy="521" r="1.6" fill="#5aa9ff" fill-opacity="0.3"/><circle cx="546.8" cy="87.2" r="1.6" fill="#ff6b35" fill-opacity="0.3"/><circle cx="276.8" cy="224.5" r="2.4" fill="#d9c7a1" fill-opacity="0.55"/><circle cx="477.1" cy="200.6" r="1.6" fill="#ff6b35" fill-opacity="0.3"/><circle cx="167.9" cy="146.2" r="2.4" fill="#d9c7a1" fill-opacity="0.55"/><circle cx="438.5" cy="110.2" r="1.6" fill="#ff6b35" fill-opacity="0.3"/><circle cx="432" cy="378.4" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="454.1" cy="555.3" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="1002.6" cy="513.2" r="2.4" fill="#ff5d8f" fill-opacity="0.55"/><circle cx="465.2" cy="160.8" r="2.4" fill="#ff6b35" fill-opacity="0.55"/><circle cx="287.4" cy="436.1" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="1118.2" cy="573.3" r="1.6" fill="#ff5d8f" fill-opacity="0.3"/><circle cx="438.6" cy="163.7" r="2.4" fill="#ff6b35" fill-opacity="0.55"/><circle cx="918.7" cy="568.7" r="1.6" fill="#ff5d8f" fill-opacity="0.3"/><circle cx="730.4" cy="399.5" r="2.4" fill="#5aa9ff" fill-opacity="0.55"/><circle cx="790" cy="522.9" r="1.6" fill="#5aa9ff" fill-opacity="0.3"/><circle cx="370.2" cy="503.9" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="499.9" cy="116" r="2.4" fill="#ff6b35" fill-opacity="0.55"/><circle cx="885.5" cy="131.7" r="1.6" fill="#ffc857" fill-opacity="0.3"/><circle cx="512.3" cy="433.9" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="1015.6" cy="244.7" r="2.4" fill="#9ad97a" fill-opacity="0.55"/><circle cx="1081.4" cy="193.5" r="2.4" fill="#9ad97a" fill-opacity="0.55"/>
34<circle cx="1075.8" cy="270.2" r="2.4" fill="#9ad97a" fill-opacity="0.55"/><circle cx="989.9" cy="186.6" r="1.6" fill="#9ad97a" fill-opacity="0.3"/><circle cx="347.3" cy="557.8" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="684.8" cy="449" r="1.6" fill="#5aa9ff" fill-opacity="0.3"/><circle cx="338.4" cy="431.8" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="364.5" cy="376.2" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="393.1" cy="427.6" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="70" cy="219.8" r="2.4" fill="#d9c7a1" fill-opacity="0.55"/><circle cx="506" cy="520.3" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="1070.8" cy="463.5" r="1.6" fill="#ff5d8f" fill-opacity="0.3"/><circle cx="278.7" cy="490.3" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="734.7" cy="205.8" r="2.4" fill="#ffc857" fill-opacity="0.55"/><circle cx="370.6" cy="465.8" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="755.9" cy="259.2" r="1.6" fill="#ffc857" fill-opacity="0.3"/><circle cx="1149.6" cy="224.1" r="2.4" fill="#9ad97a" fill-opacity="0.55"/><circle cx="472.7" cy="389" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="545.8" cy="157.1" r="2.4" fill="#ff6b35" fill-opacity="0.55"/><circle cx="92.6" cy="545.7" r="1.6" fill="#7b6cff" fill-opacity="0.3"/><circle cx="483.8" cy="233.1" r="1.6" fill="#ff6b35" fill-opacity="0.3"/><circle cx="1047.3" cy="624.5" r="1.6" fill="#ff5d8f" fill-opacity="0.3"/><circle cx="424.3" cy="181" r="1.6" fill="#ff6b35" fill-opacity="0.3"/><circle cx="414.9" cy="569.6" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="427.6" cy="455.9" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="406.3" cy="90.2" r="1.6" fill="#ff6b35" fill-opacity="0.3"/><circle cx="998.1" cy="283.6" r="2.4" fill="#9ad97a" fill-opacity="0.55"/><circle cx="304.4" cy="404.2" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="526.8" cy="466.8" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="639.9" cy="463.3" r="1.6" fill="#5aa9ff" fill-opacity="0.3"/><circle cx="803" cy="425.7" r="2.4" fill="#5aa9ff" fill-opacity="0.55"/><circle cx="227.7" cy="110.7" r="2.4" fill="#d9c7a1" fill-opacity="0.55"/><circle cx="308.7" cy="541.3" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="743.7" cy="93.8" r="1.6" fill="#ffc857" fill-opacity="0.3"/><circle cx="407.2" cy="367.6" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="585.8" cy="154.3" r="1.6" fill="#ff6b35" fill-opacity="0.3"/><circle cx="893.5" cy="210.1" r="1.6" fill="#ffc857" fill-opacity="0.3"/><circle cx="713.4" cy="543" r="1.6" fill="#5aa9ff" fill-opacity="0.3"/><circle cx="390.8" cy="497.3" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="680" cy="406.4" r="1.6" fill="#5aa9ff" fill-opacity="0.3"/><circle cx="218.6" cy="219.4" r="2.4" fill="#d9c7a1" fill-opacity="0.55"/><circle cx="397.6" cy="204.4" r="1.6" fill="#ff6b35" fill-opacity="0.3"/><circle cx="203.3" cy="277.6" r="1.6" fill="#d9c7a1" fill-opacity="0.3"/><circle cx="481.6" cy="549.8" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="271.7" cy="455.2" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="683.2" cy="205.3" r="1.6" fill="#ffc857" fill-opacity="0.3"/><circle cx="927.8" cy="492.1" r="1.6" fill="#ff5d8f" fill-opacity="0.3"/><circle cx="478.8" cy="464.8" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="843.6" cy="94.4" r="1.6" fill="#ffc857" fill-opacity="0.3"/><circle cx="437.7" cy="481" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="345.5" cy="517.1" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="382.6" cy="443.2" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="821.6" cy="490.1" r="1.6" fill="#5aa9ff" fill-opacity="0.3"/><circle cx="507.7" cy="411.3" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="370" cy="572" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="1129.2" cy="525.4" r="1.6" fill="#ff5d8f" fill-opacity="0.3"/><circle cx="79.5" cy="140.3" r="1.6" fill="#d9c7a1" fill-opacity="0.3"/><circle cx="400.4" cy="404.9" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="455.4" cy="518.8" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="335.7" cy="378.1" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="951.3" cy="610.3" r="1.6" fill="#ff5d8f" fill-opacity="0.3"/><circle cx="1044.2" cy="241.6" r="2.4" fill="#9ad97a" fill-opacity="0.55"/><circle cx="644.5" cy="505.1" r="2.4" fill="#5aa9ff" fill-opacity="0.55"/><circle cx="804.1" cy="175" r="2.4" fill="#ffc857" fill-opacity="0.55"/><circle cx="818.8" cy="264.1" r="1.6" fill="#ffc857" fill-opacity="0.3"/><circle cx="1062.3" cy="209" r="2.4" fill="#9ad97a" fill-opacity="0.55"/><circle cx="494.9" cy="64.6" r="1.6" fill="#ff6b35" fill-opacity="0.3"/><circle cx="1060.5" cy="156" r="2.4" fill="#9ad97a" fill-opacity="0.55"/><circle cx="1123" cy="285.5" r="2.4" fill="#9ad97a" fill-opacity="0.55"/><circle cx="763.8" cy="397.1" r="2.4" fill="#5aa9ff" fill-opacity="0.55"/><circle cx="541.6" cy="221.6" r="1.6" fill="#ff6b35" fill-opacity="0.3"/><circle cx="805.1" cy="121.3" r="2.4" fill="#ffc857" fill-opacity="0.55"/><circle cx="373.2" cy="130.3" r="2.4" fill="#ff6b35" fill-opacity="0.55"/><circle cx="525.7" cy="503.2" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="156.2" cy="530" r="2.4" fill="#7b6cff" fill-opacity="0.55"/><circle cx="240.2" cy="165.9" r="1.6" fill="#d9c7a1" fill-opacity="0.3"/><circle cx="841.3" cy="166" r="2.4" fill="#ffc857" fill-opacity="0.55"/><circle cx="288.2" cy="174.9" r="1.6" fill="#d9c7a1" fill-opacity="0.3"/><circle cx="767" cy="542.8" r="1.6" fill="#5aa9ff" fill-opacity="0.3"/><circle cx="278.7" cy="513.7" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="452.8" cy="371.8" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="640.3" cy="436" r="2.4" fill="#5aa9ff" fill-opacity="0.55"/><circle cx="692.8" cy="124.5" r="1.6" fill="#ffc857" fill-opacity="0.3"/><circle cx="103.8" cy="262.8" r="1.6" fill="#d9c7a1" fill-opacity="0.3"/><circle cx="824.9" cy="223.8" r="2.4" fill="#ffc857" fill-opacity="0.55"/><circle cx="494.7" cy="187.4" r="2.4" fill="#ff6b35" fill-opacity="0.55"/><circle cx="576.2" cy="106.4" r="1.6" fill="#ff6b35" fill-opacity="0.3"/><circle cx="482.1" cy="130.5" r="2.4" fill="#ff6b35" fill-opacity="0.55"/><circle cx="281.1" cy="418.8" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="444.2" cy="571.5" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="162.8" cy="491" r="1.6" fill="#7b6cff" fill-opacity="0.3"/><circle cx="531.5" cy="443.4" r="1.6" fill="#3de0c5" fill-opacity="0.3"/><circle cx="445.6" cy="234.8" r="1.6" fill="#ff6b35" fill-opacity="0.3"/><circle cx="325.2" cy="561.2" r="2.4" fill="#3de0c5" fill-opacity="0.55"/><circle cx="825.9" cy="446.3" r="2.4" fill="#5aa9ff" fill-opacity="0.55"/><circle cx="905.1" cy="159.6" r="1.6" fill="#ffc857" fill-opacity="0.3"/><a href="/guide/activation-function" class="fg-map__mark"><title>Activation Function: A non-linear function applied to neuron outputs that introduces non-linearity, enabling networks to learn complex patterns.</title><circle cx="479.9" cy="199.9" r="9" fill="#ff6b35" fill-opacity=".18"/><circle cx="479.9" cy="199.9" r="5" fill="#ff6b35"/></a><a href="/guide/accuracy" class="fg-map__mark"><title>Accuracy: The share of predictions a classifier gets right. Simple and intuitive, but misleading when one class is much rarer than the others.</title><circle cx="1007.7" cy="224.7" r="9" fill="#9ad97a" fill-opacity=".18"/><circle cx="1007.7" cy="224.7" r="5" fill="#9ad97a"/></a><a href="/guide/ai-agent" class="fg-map__mark"><title>AI Agent: A system where a large language model decides its own next steps in a loop: calling tools, reading the results, and continuing until the task is done.</title><circle cx="109.9" cy="501.3" r="9" fill="#7b6cff" fill-opacity=".18"/><circle cx="109.9" cy="501.3" r="5" fill="#7b6cff"/></a><a href="/guide/agent" class="fg-map__mark"><title>Agent: In reinforcement learning, the learner and decision-maker that observes an environment, takes actions and adjusts its behaviour to maximize cumulative reward.</title><circle cx="123.9" cy="565.2" r="9" fill="#7b6cff" fill-opacity=".18"/><circle cx="123.9" cy="565.2" r="5" fill="#7b6cff"/></a><a href="/guide/attention-mechanism" class="fg-map__mark"><title>Attention Mechanism: A technique that lets a neural network weigh every part of its input when producing each output, focusing on the parts most relevant at that step.</title><circle cx="407.5" cy="530.6" r="9" fill="#3de0c5" fill-opacity=".18"/><circle cx="407.5" cy="530.6" r="5" fill="#3de0c5"/></a><a href="/guide/backpropagation" class="fg-map__mark"><title>Backpropagation: The algorithm for computing gradients of the loss with respect to network weights, enabling training through gradient descent.</title><circle cx="751.2" cy="217" r="9" fill="#ffc857" fill-opacity=".18"/><circle cx="751.2" cy="217" r="5" fill="#ffc857"/></a><a href="/guide/classification" class="fg-map__mark"><title>Classification: A supervised learning task where the model assigns each input to one of a fixed set of categories, such as spam or not spam.</title><circle cx="235.8" cy="207.7" r="9" fill="#d9c7a1" fill-opacity=".18"/><circle cx="235.8" cy="207.7" r="5" fill="#d9c7a1"/><text x="244.8" y="211.7" text-anchor="start">Classification</text></a><a href="/guide/cnn" class="fg-map__mark"><title>Convolutional Neural Network: A neural network that scans images with small learned filters, reusing the same weights at every position to build up from edges to whole objects.</title>
34<circle cx="534.1" cy="125.1" r="9" fill="#ff6b35" fill-opacity=".18"/><circle cx="534.1" cy="125.1" r="5" fill="#ff6b35"/><text x="543.1" y="129.1" text-anchor="start">Convolutional Neural Network</text></a><a href="/guide/computer-vision" class="fg-map__mark"><title>Computer Vision: The field of AI that gets computers to extract meaning from images and video: what is in them, where it is, and how it moves.</title><circle cx="749.3" cy="427.7" r="9" fill="#5aa9ff" fill-opacity=".18"/><circle cx="749.3" cy="427.7" r="5" fill="#5aa9ff"/><text x="758.3" y="431.7" text-anchor="start">Computer Vision</text></a><a href="/guide/context-window" class="fg-map__mark"><title>Context Window: The maximum number of tokens a language model can take into account at once, counting both the input it reads and the output it writes. Also called context length.</title><circle cx="324.5" cy="462.4" r="9" fill="#3de0c5" fill-opacity=".18"/><circle cx="324.5" cy="462.4" r="5" fill="#3de0c5"/></a><a href="/guide/deep-learning" class="fg-map__mark"><title>Deep Learning: A subset of machine learning that uses neural networks with multiple layers (deep neural networks) to learn hierarchical representations of data.</title><circle cx="439.8" cy="188.2" r="9" fill="#ff6b35" fill-opacity=".18"/><circle cx="439.8" cy="188.2" r="5" fill="#ff6b35"/></a><a href="/guide/diffusion-model" class="fg-map__mark"><title>Diffusion Model: A generative model trained to remove noise a step at a time, so it can turn pure random noise into a new image, audio clip or other sample.</title><circle cx="674.5" cy="477.8" r="9" fill="#5aa9ff" fill-opacity=".18"/><circle cx="674.5" cy="477.8" r="5" fill="#5aa9ff"/></a><a href="/guide/dropout" class="fg-map__mark"><title>Dropout: A regularization technique that randomly switches off units during training so the network can't lean on any one of them, which reduces overfitting.</title><circle cx="458.7" cy="103.1" r="9" fill="#ff6b35" fill-opacity=".18"/><circle cx="458.7" cy="103.1" r="5" fill="#ff6b35"/></a><a href="/guide/embedding" class="fg-map__mark"><title>Embedding: A list of numbers (a vector) that represents a word, sentence, image or other item, learned so that similar items end up close together.</title><circle cx="454.5" cy="427.5" r="9" fill="#3de0c5" fill-opacity=".18"/><circle cx="454.5" cy="427.5" r="5" fill="#3de0c5"/><text x="463.5" y="431.5" text-anchor="start">Embedding</text></a><a href="/guide/false-positive" class="fg-map__mark"><title>False Positive: A case the model labels positive that is actually negative, such as a legitimate email sent to spam. Statisticians call it a Type I error.</title><circle cx="1110" cy="216.6" r="9" fill="#9ad97a" fill-opacity=".18"/><circle cx="1110" cy="216.6" r="5" fill="#9ad97a"/><text x="1101" y="220.6" text-anchor="end">False Positive</text></a><a href="/guide/feature" class="fg-map__mark"><title>Feature: A single measurable property of an example, such as a house'
34s floor area or how many links an email contains, used as an input to a model.</title><circle cx="202.3" cy="236.9" r="9" fill="#d9c7a1" fill-opacity=".18"/><circle cx="202.3" cy="236.9" r="5" fill="#d9c7a1"/></a><a href="/guide/fine-tuning" class="fg-map__mark"><title>Fine-Tuning: The process of further training a pre-trained model on a specific dataset to adapt it for a particular task or domain.</title><circle cx="785.8" cy="227.1" r="9" fill="#ffc857" fill-opacity=".18"/><circle cx="785.8" cy="227.1" r="5" fill="#ffc857"/></a><a href="/guide/foundation-model" class="fg-map__mark"><title>Foundation Model: A large model pre-trained on broad data at scale that can be adapted to many downstream tasks, such as GPT-style LLMs, BERT or CLIP.</title><circle cx="392.5" cy="409.4" r="9" fill="#3de0c5" fill-opacity=".18"/><circle cx="392.5" cy="409.4" r="5" fill="#3de0c5"/></a><a href="/guide/generalization" class="fg-map__mark"><title>Generalization: A model's ability to perform well on new data it never saw during training, which is the whole point of learning from examples.</title><circle cx="147.7" cy="143.1" r="9" fill="#d9c7a1" fill-opacity=".18"/><circle cx="147.7" cy="143.1" r="5" fill="#d9c7a1"/></a><a href="/guide/gradient-descent" class="fg-map__mark"><title>Gradient Descent: An optimization method that repeatedly moves a model's parameters a small step in the direction that most reduces the loss.</title><circle cx="842" cy="143.4" r="9" fill="#ffc857" fill-opacity=".18"/><circle cx="842" cy="143.4" r="5" fill="#ffc857"/><text x="851" y="147.4" text-anchor="start">Gradient Descent</text></a><a href="/guide/hallucination" class="fg-map__mark"><title>Hallucination: When a language model produces fluent, confident output that is false or unsupported by its sources, such as invented facts, quotes or citations.</title><circle cx="336.1" cy="436.9" r="9" fill="#3de0c5" fill-opacity=".18"/><circle cx="336.1" cy="436.9" r="5" fill="#3de0c5"/></a><a href="/guide/hyperparameter" class="fg-map__mark"><title>Hyperparameter: A setting chosen before training, such as the learning rate or batch size, that controls how a model learns but isn't itself learned from the data.</title><circle cx="743.5" cy="138.3" r="9" fill="#ffc857" fill-opacity=".18"/><circle cx="743.5" cy="138.3" r="5" fill="#ffc857"/></a><a href="/guide/image-classification" class="fg-map__mark"><title>Image Classification: Assigning one label from a fixed set of categories to a whole image, such as "cat" or "defective part"; a core computer vision task.</title><circle cx="766.2" cy="504.6" r="9" fill="#5aa9ff" fill-opacity=".18"/><circle cx="766.2" cy="504.6" r="5" fill="#5aa9ff"/><text x="775.2" y="508.6" text-anchor="start">Image Classification</text></a><a href="/guide/image-generation" class="fg-map__mark"><title>Image Generation: Creating new images, often from a text prompt, with generative models such as GANs, VAEs and diffusion models that learn to turn random noise into realistic pictures.</title><circle cx="710.7" cy="512.3" r="9" fill="#5aa9ff" fill-opacity=".18"/><circle cx="710.7" cy="512.3" r="5" fill="#5aa9ff"/></a><a href="/guide/inference" class="fg-map__mark"><title>
34Inference: Running a trained model on new inputs to get predictions, with its weights frozen: the stage of a model's life that users actually interact with.</title><circle cx="1041.4" cy="493.1" r="9" fill="#ff5d8f" fill-opacity=".18"/><circle cx="1041.4" cy="493.1" r="5" fill="#ff5d8f"/><text x="1050.4" y="497.1" text-anchor="start">Inference</text></a><a href="/guide/iou" class="fg-map__mark"><title>Intersection over Union: A score from 0 to 1 for how well a predicted region matches the true one: the area they share divided by the area they cover together.</title><circle cx="1032.5" cy="194" r="9" fill="#9ad97a" fill-opacity=".18"/><circle cx="1032.5" cy="194" r="5" fill="#9ad97a"/></a><a href="/guide/language-modeling" class="fg-map__mark"><title>Language Modeling: Assigning probabilities to sequences of text, in practice by predicting each next token from the ones before it; the objective behind large language models.</title><circle cx="463.9" cy="503.1" r="9" fill="#3de0c5" fill-opacity=".18"/><circle cx="463.9" cy="503.1" r="5" fill="#3de0c5"/></a><a href="/guide/learning-rate" class="fg-map__mark"><title>Learning Rate: The hyperparameter that sets how big a step gradient descent takes on each update; too high makes training unstable, too low makes it crawl.</title><circle cx="729" cy="193.6" r="9" fill="#ffc857" fill-opacity=".18"/><circle cx="729" cy="193.6" r="5" fill="#ffc857"/></a><a href="/guide/llm" class="fg-map__mark"><title>Large Language Model: A neural network, almost always a transformer, trained on vast amounts of text to predict the next token, which lets it write, answer, summarize and follow instructions.</title><circle cx="426.1" cy="412.8" r="9" fill="#3de0c5" fill-opacity=".18"/><circle cx="426.1" cy="412.8" r="5" fill="#3de0c5"/><text x="435.1" y="416.8" text-anchor="start">Large Language Model</text></a><a href="/guide/loss-function" class="fg-map__mark"><title>Loss Function: A function that scores how wrong a model's prediction is as a single number, which training then works to make as small as possible.</title><circle cx="855.1" cy="170.9" r="9" fill="#ffc857" fill-opacity=".18"/><circle cx="855.1" cy="170.9" r="5" fill="#ffc857"/><text x="864.1" y="174.9" text-anchor="start">Loss Function</text></a><a href="/guide/machine-learning" class="fg-map__mark"><title>Machine Learning: Building systems that learn patterns from data instead of following hand-written rules, getting better at a task as they see more examples.</title><circle cx="116.4" cy="211.8" r="9" fill="#d9c7a1" fill-opacity=".18"/><circle cx="116.4" cy="211.8" r="5" fill="#d9c7a1"/></a><a href="/guide/neural-network" class="fg-map__mark"><title>Neural Network: A model built from layers of simple units, each taking a weighted sum of its inputs and applying a nonlinearity; training adjusts the weights so the whole stack maps inputs to useful outputs.</title><circle cx="501.4" cy="103.1" r="9" fill="#ff6b35" fill-opacity=".18"/><circle cx="501.4" cy="103.1" r="5" fill="#ff6b35"/><text x="510.4" y="107.1" text-anchor="start">Neural Network</text></a><a href="/guide/nlp" class="fg-map__mark"><title>Natural Language Processing: The field of AI that lets computers read, interpret, translate and generate human language, from spam filters and search to chatbots.</title><circle cx="472.2" cy="450.7" r="9" fill="#3de0c5" fill-opacity=".18"/><circle cx="472.2" cy="450.7" r="5" fill="#3de0c5"/><text x="481.2" y="454.7" text-anchor="start">Natural Language Processing</text></a><a href="/guide/object-detection" class="fg-map__mark"><title>Object Detection: Finding every object of interest in an image and giving each a class label, a confidence score and a bounding box.</title><circle cx="785.5" cy="462.2" r="9" fill="#5aa9ff" fill-opacity=".18"/><circle cx="785.5" cy="462.2" r="5" fill="#5aa9ff"/><text x="794.5" y="466.2" text-anchor="start">Object Detection</text></a><a href="/guide/overfitting" class="fg-map__mark"><title>Overfitting: When a model fits its training data too closely, noise included, so it scores well on examples it has seen and poorly on new ones.</title><circle cx="847.5" cy="199.7" r="9" fill="#ffc857" fill-opacity=".18"/><circle cx="847.5" cy="199.7" r="5" fill="#ffc857"/></a><a href="/guide/parameter" class="fg-map__mark"><title>Parameter: A number inside a model, a weight or a bias, that training adjusts to reduce the loss. Together, a model's parameters are everything it has learned.</title><circle cx="418.5" cy="158.6" r="9" fill="#ff6b35" fill-opacity=".18"/><circle cx="418.5" cy="158.6" r="5" fill="#ff6b35"/></a><a href="/guide/policy" class="fg-map__mark"><title>Policy: The rule an RL agent uses to choose actions: a mapping from each state to an action, or to a probability distribution over actions.</title><circle cx="186.1" cy="523.5" r="9" fill="#7b6cff" fill-opacity=".18"/><circle cx="186.1" cy="523.5" r="5" fill="#7b6cff"/><text x="195.1" y="527.5" text-anchor="start">Policy</text></a><a href="/guide/pre-training" class="fg-map__mark"><title>Pre-training: Training a model on a large dataset (often self-supervised) before fine-tuning on specific tasks, enabling transfer learning.</title><circle cx="821.7" cy="220.7" r="9" fill="#ffc857" fill-opacity=".18"/><circle cx="821.7" cy="220.7" r="5" fill="#ffc857"/></a><a href="/guide/precision" class="fg-map__mark"><title>Precision: Of everything a model flagged as positive, the fraction that really was positive: TP / (TP + FP). It answers "when the model says yes, how often is it right?"</title><circle cx="1104.3" cy="252.9" r="9" fill="#9ad97a" fill-opacity=".18"/><circle cx="1104.3" cy="252.9" r="5" fill="#9ad97a"/><text x="1113.3" y="256.9" text-anchor="start">Precision</text></a><a href="/guide/prompt-engineering" class="fg-map__mark"><title>Prompt Engineering: Designing a model's input (instructions, context, examples and output format) to get better results without changing the model itself.</title><circle cx="327.8" cy="489.3" r="9" fill="#3de0c5" fill-opacity=".18"/><circle cx="327.8" cy="489.3" r="5" fill="#3de0c5"/></a><a href="/guide/rag" class="fg-map__mark"><title>Retrieval-Augmented Generation: Fetching relevant documents at question time and adding them to an LLM's prompt, so answers can draw on current, checkable sources without retraining.</title><circle cx="360.4" cy="418" r="9" fill="#3de0c5" fill-opacity=".18"/><circle cx="360.4" cy="418" r="5" fill="#3de0c5"/></a><a href="/guide/recall" class="fg-map__mark"><title>Recall: The share of actual positives a model correctly identifies, TP / (TP + FN); also called sensitivity or the true positive rate.</title>
34<circle cx="1078.1" cy="190.4" r="9" fill="#9ad97a" fill-opacity=".18"/><circle cx="1078.1" cy="190.4" r="5" fill="#9ad97a"/><text x="1087.1" y="194.4" text-anchor="start">Recall</text></a><a href="/guide/regression" class="fg-map__mark"><title>Regression: A supervised learning task where the model predicts continuous numerical values rather than discrete categories.</title><circle cx="152.8" cy="238.6" r="9" fill="#d9c7a1" fill-opacity=".18"/><circle cx="152.8" cy="238.6" r="5" fill="#d9c7a1"/></a><a href="/guide/regularization" class="fg-map__mark"><title>Regularization: Any change to training meant to make a model generalize better rather than fit its training data more closely, such as weight penalties, dropout or early stopping.</title><circle cx="726.1" cy="164.3" r="9" fill="#ffc857" fill-opacity=".18"/><circle cx="726.1" cy="164.3" r="5" fill="#ffc857"/></a><a href="/guide/reinforcement-learning" class="fg-map__mark"><title>Reinforcement Learning: Learning through interaction with an environment, receiving rewards or penalties to learn optimal behavior policies.</title><circle cx="156.1" cy="494.8" r="9" fill="#7b6cff" fill-opacity=".18"/><circle cx="156.1" cy="494.8" r="5" fill="#7b6cff"/><text x="165.1" y="498.8" text-anchor="start">Reinforcement Learning</text></a><a href="/guide/representation-learning" class="fg-map__mark"><title>Representation Learning: Letting a model learn its own internal description of the data, such as embeddings or layered visual features, instead of relying on features designed by hand.</title><circle cx="426" cy="125" r="9" fill="#ff6b35" fill-opacity=".18"/><circle cx="426" cy="125" r="5" fill="#ff6b35"/></a><a href="/guide/reward" class="fg-map__mark"><title>Reward: The single number an environment sends back after each action, telling a reinforcement learning agent how good that step was; the agent learns to maximize its long-run total.</title><circle cx="170.1" cy="558.7" r="9" fill="#7b6cff" fill-opacity=".18"/><circle cx="170.1" cy="558.7" r="5" fill="#7b6cff"/><text x="179.1" y="562.7" text-anchor="start">Reward</text></a><a href="/guide/rnn" class="fg-map__mark"><title>Recurrent Neural Network: A neural network architecture with loops that allow information to persist, designed for sequential data like text and time series.</title><circle cx="520.1" cy="188.3" r="9" fill="#ff6b35" fill-opacity=".18"/><circle cx="520.1" cy="188.3" r="5" fill="#ff6b35"/></a><a href="/guide/self-supervised-learning" class="fg-map__mark"><title>Self-Supervised Learning: Training on unlabeled data by making the data supply its own answers: hide or alter part of an input and have the model predict or match it.</title><circle cx="775.7" cy="124" r="9" fill="#ffc857" fill-opacity=".18"/><circle cx="775.7" cy="124" r="5" fill="#ffc857"/></a><a href="/guide/semantic-segmentation" class="fg-map__mark"><title>Semantic Segmentation: Labelling every pixel in an image with a class, such as road, car or sky, so the output is a map of what is where rather than a single tag.</title><circle cx="693.8" cy="435.4" r="9" fill="#5aa9ff" fill-opacity=".18"/><circle cx="693.8" cy="435.4" r="5" fill="#5aa9ff"/></a><a href="/guide/self-attention" class="fg-map__mark"><title>Self-Attention: An attention step in which every token in a sequence looks at every token in that same sequence and builds a new representation from a weighted mix of them.</title><circle cx="345.5" cy="512.5" r="9" fill="#3de0c5" fill-opacity=".18"/><circle cx="345.5" cy="512.5" r="5" fill="#3de0c5"/></a><a href="/guide/softmax" class="fg-map__mark"><title>Softmax: A function that turns a list of scores (logits) into probabilities that are all positive and sum to 1; the standard output of classifiers and language models.</title><circle cx="541.4" cy="158.7" r="9" fill="#ff6b35" fill-opacity=".18"/><circle cx="541.4" cy="158.7" r="5" fill="#ff6b35"/><text x="550.4" y="162.7" text-anchor="start">Softmax</text></a><a href="/guide/supervised-learning" class="fg-map__mark"><title>Supervised Learning: Learning from examples paired with the correct answer, so a model can predict answers for new input
34s it hasn't seen.</title><circle cx="197.2" cy="141.4" r="9" fill="#d9c7a1" fill-opacity=".18"/><circle cx="197.2" cy="141.4" r="5" fill="#d9c7a1"/><text x="206.2" y="145.4" text-anchor="start">Supervised Learning</text></a><a href="/guide/test-set" class="fg-map__mark"><title>Test Set: Data locked away until the end of a project and used once to estimate how the finished model will perform on new, unseen examples.</title><circle cx="1065.2" cy="272" r="9" fill="#9ad97a" fill-opacity=".18"/><circle cx="1065.2" cy="272" r="5" fill="#9ad97a"/></a><a href="/guide/tokenization" class="fg-map__mark"><title>Tokenization: Splitting text into tokens, usually subword pieces, and mapping each to an integer ID so a language model can process it.</title><circle cx="439.6" cy="522" r="9" fill="#3de0c5" fill-opacity=".18"/><circle cx="439.6" cy="522" r="5" fill="#3de0c5"/></a><a href="/guide/token" class="fg-map__mark"><title>Token: The unit of text a language model reads and writes: a word, part of a word, or a character, drawn from the model's fixed vocabulary.</title><circle cx="373.9" cy="527.2" r="9" fill="#3de0c5" fill-opacity=".18"/><circle cx="373.9" cy="527.2" r="5" fill="#3de0c5"/></a><a href="/guide/tool-use" class="fg-map__mark"><title>Tool Use: A language model asking the application around it to run a function, such as search, a calculator or an API, then using the result in its answer.</title><circle cx="93.9" cy="536.5" r="9" fill="#7b6cff" fill-opacity=".18"/><circle cx="93.9" cy="536.5" r="5" fill="#7b6cff"/></a><a href="/guide/training-data" class="fg-map__mark"><title>Training Data: The examples a model learns its weights from, kept separate from the validation and test data used to check how well it generalizes.</title><circle cx="233.6" cy="168.2" r="9" fill="#d9c7a1" fill-opacity=".18"/><circle cx="233.6" cy="168.2" r="5" fill="#d9c7a1"/><text x="242.6" y="172.2" text-anchor="start">Training Data</text></a><a href="/guide/training" class="fg-map__mark"><title>Training: The process of fitting a model to data by repeatedly measuring how wrong its outputs are and adjusting its parameters to reduce that error.</title><circle cx="812.4" cy="125.9" r="9" fill="#ffc857" fill-opacity=".18"/><circle cx="812.4" cy="125.9" r="5" fill="#ffc857"/><text x="821.4" y="129.9" text-anchor="start">Training</text></a><a href="/guide/transformer" class="fg-map__mark"><title>Transformer: A neural network architecture, introduced in 2017, built from stacked self-attention and feed-forward layers; the basis of nearly every modern large language model.</title><circle cx="475.5" cy="477.6" r="9" fill="#3de0c5" fill-opacity=".18"/><circle cx="475.5" cy="477.6" r="5" fill="#3de0c5"/></a><a href="/guide/unsupervised-learning" class="fg-map__mark"><title>Unsupervised Learning: Learning from unlabeled data to discover hidden patterns, structures, or relationships without explicit target outputs.</title><circle cx="114.2" cy="172.3" r="9" fill="#d9c7a1" fill-opacity=".18"/><circle cx="114.2" cy="172.3" r="5" fill="#d9c7a1"/></a><a href="/guide/validation-set" class="fg-map__mark"><title>Validation Set: Data held back from training and used to choose hyperparameters and checkpoints, so those choices are judged on examples the model never fit.</title><circle cx="1022.2" cy="259.4" r="9" fill="#9ad97a" fill-opacity=".18"/><circle cx="1022.2" cy="259.4" r="5" fill="#9ad97a"/></a></svg></div> <!-- Phones: regions as chips instead of the map. --> <nav class="fg-region-chips mt-8 flex flex-wrap gap-2" aria-label="Regions"> <a href="#region-foundations" class="fg-tier" style="--rc: #d9c7a1"> <span class="fg-dot"></span> Foundations · 78 </a><a href="#region-neural-nets" class="fg-tier" style="--rc: #ff6b35"> <span class="fg-dot"></span> Neural Networks · 71 </a><a href="#region-training" class="fg-tier" style="--rc: #ffc857"> <span class="fg-dot"></span> Training · 80 </a><a href="#region-evaluation" class="fg-tier" style="--rc: #9ad97a"> <span class="fg-dot"></span> Evaluation · 45 </a><a href="#region-language" class="fg-tier" style="--rc: #3de0c5"> <span class="fg-dot"></span> Language & LLMs · 118 </a>
34<a href="#region-vision" class="fg-tier" style="--rc: #5aa9ff"> <span class="fg-dot"></span> Vision & Multimodal · 54 </a><a href="#region-agents" class="fg-tier" style="--rc: #7b6cff"> <span class="fg-dot"></span> Agents & RL · 32 </a><a href="#region-shipping" class="fg-tier" style="--rc: #ff5d8f"> <span class="fg-dot"></span> Shipping AI · 71 </a> </nav> </div> </section> <div class="mx-auto flex max-w-[76rem] flex-col gap-14 px-4 py-16 sm:px-8"> <p class="max-w-3xl text-muted-foreground"> <strong class="text-foreground">How to use it:</strong> open any term and its 35<em>route</em> shows every idea to understand first, basics first. <strong class="text-foreground">Landmarks</strong> 36are the concepts the rest of the map leans on, which makes them good first stops. 37</p> <section id="region-foundations" class="fg-region" style="--rc: #d9c7a1" aria-labelledby="h-foundations"> <div class="flex flex-wrap items-baseline justify-between gap-3"> <h2 id="h-foundations" class="fg-display flex items-center gap-3 text-3xl text-foreground"> <span class="fg-dot" style="width: 0.85rem; height: 0.85rem"></span> Foundations </h2> <p class="fg-mono text-muted-foreground"> 78 terms · 8 landmarks 38</p> </div> <p class="mt-2 max-w-2xl text-muted-foreground">Core ML ideas, classic algorithms, and the data they learn from.</p> <div class="mt-6 grid gap-3 sm:grid-cols-2 lg:grid-cols-4"> <a href="/guide/supervised-learning" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 11</span> <span class="fg-display text-lg text-foreground">Supervised Learning</span> <span class="line-clamp-2 text-sm text-muted-foreground">Learning from examples paired with the correct answer, so a model can predict answers for new inputs it hasn't seen.</span> </a><a href="/guide/training-data" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 11</span> <span class="fg-display text-lg text-foreground">Training Data</span> <span class="line-clamp-2 text-sm text-muted-foreground">The examples a model learns its weights from, kept separate from the validation and test data used to check how well it generalizes.</span> </a><a href="/guide/classification" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 10</span> <span class="fg-display text-lg text-foreground">Classification</span> <span class="line-clamp-2 text-sm text-muted-foreground">A supervised learning task where the model assigns each input to one of a fixed set of categories, such as spam or not spam.</span> </a><a href="/guide/feature" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 10</span> <span class="fg-display text-lg text-foreground">Feature</span> <span class="line-clamp-2 text-sm text-muted-foreground">A single measurable property of an example, such as a house's floor area or how many links an email contains, used as an input to a model.</span> </a><a href="/guide/regression" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 7</span> <span class="fg-display text-lg text-foreground">Regression</span> <span class="line-clamp-2 text-sm text-muted-foreground">A supervised learning task where the model predicts continuous numerical values rather than discrete categories.</span> </a><a href="/guide/machine-learning" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 6</span> <span class="fg-display text-lg text-foreground">Machine Learning</span> <span class="line-clamp-2 text-sm text-muted-foreground">Building systems that learn patterns from data instead of following hand-written rules, getting better at a task as they see more examples.</span> </a><a href="/guide/unsupervised-learning" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 6</span> <span class="fg-display text-lg text-foreground">Unsupervised Learning</span> <span class="line-clamp-2 text-sm text-muted-foreground">Learning from unlabeled data to discover hidden patterns, structures, or relationships without explicit target outputs.</span> </a><a href="/guide/generalization" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 0</span> <span class="fg-display text-lg text-foreground">Generalization</span> <span class="line-clamp-2 text-sm text-muted-foreground">A model's ability to perform well on new data it never saw during training, which is the whole point of learning from examples.</span> </a> </div> <details class="fg-all mt-6"> <summary>All 78 Foundations terms</summary> <ul class="fg-terms mt-4"> <li> <a href="/guide/active-learning" data-tier="3"> Active Learning </a> </li><li> <a href="/guide/alphafold" data-tier="3"> AlphaFold </a> </li><li> <a href="/guide/anomaly-detection" data-tier="2"> Anomaly Detection </a> </li><li> <a href="/guide/bagging" data-tier="2"> Bagging </a> </li><li> <a href="/guide/bayesian-inference" data-tier="2"> Bayesian Inference </a> </li><li> <a href="/guide/bias-variance-tradeoff" data-tier="2"> Bias-Variance Tradeoff </a> </li><li> <a href="/guide/causal-inference" data-tier="3"> Causal Inference </a> </li><li> <a href="/guide/classification" data-tier="1"> Classification </a> </li><li> <a href="/guide/clustering" data-tier="2"> Clustering </a> </li><li> <a href="/guide/collaborative-filtering" data-tier="3"> Collaborative Filtering </a> </li><li> <a href="/guide/concentration-of-measure" data-tier="3"> Concentration of Measure </a> </li><li> <a href="/guide/conjugate-prior" data-tier="3"> Conjugate Prior </a> </li><li> <a href="/guide/cov" data-tier="3"> Covariance </a> </li><li> <a href="/guide/curse-of-dimensionality" data-tier="2"> Curse of Dimensionality </a> </li><li> <a href="/guide/data-annotation" data-tier="2"> Data Annotation </a> </li><li> <a href="/guide/data-cleaning" data-tier="2"> Data Cleaning </a> </li><li> <a href="/guide/data-preprocessing" data-tier="2"> Data Preprocessing </a> </li><li> <a href="/guide/dataset" data-tier="2"> Dataset </a> </li><li> <a href="/guide/decision-tree" data-tier="2"> Decision Tree </a> </li><li> <a href="/guide/dimensionality-reduction" data-tier="2"> Dimensionality Reduction </a> </li><li> <a href="/guide/drug-discovery" data-tier="3">
38 Drug Discovery </a> </li><li> <a href="/guide/ensemble-learning" data-tier="2"> Ensemble Learning </a> </li><li> <a href="/guide/entropy" data-tier="2"> Entropy </a> </li><li> <a href="/guide/elbo" data-tier="3"> Evidence Lower Bound </a> </li><li> <a href="/guide/em-algorithm" data-tier="3"> Expectation-Maximization </a> </li><li> <a href="/guide/feature" data-tier="1"> Feature </a> </li><li> <a href="/guide/feature-engineering" data-tier="2"> Feature Engineering </a> </li><li> <a href="/guide/feature-importance" data-tier="3"> Feature Importance </a> </li><li> <a href="/guide/feature-selection" data-tier="2"> Feature Selection </a> </li><li> <a href="/guide/fraud-detection" data-tier="3"> Fraud Detection </a> </li><li> <a href="/guide/gaussian-process" data-tier="3"> Gaussian Process </a> </li><li> <a href="/guide/generalization" data-tier="1"> Generalization </a> </li><li> <a href="/guide/gibbs-sampling" data-tier="3"> Gibbs Sampling </a> </li><li> <a href="/guide/gradient-boosting" data-tier="2"> Gradient Boosting </a> </li><li> <a href="/guide/ground-truth" data-tier="3"> Ground Truth </a> </li><li> <a href="/guide/hmm" data-tier="3"> Hidden Markov Model </a> </li><li> <a href="/guide/imbalanced-dataset" data-tier="3"> Imbalanced Dataset </a> </li><li> <a href="/guide/inductive-bias" data-tier="3"> Inductive Bias </a> </li><li> <a href="/guide/information-bottleneck" data-tier="3"> Information Bottleneck </a> </li><li> <a href="/guide/information-theory" data-tier="2"> Information Theory </a> </li><li> <a href="/guide/js-divergence" data-tier="3"> Jensen-Shannon Divergence </a> </li><li> <a href="/guide/kernel-trick" data-tier="2"> Kernel Trick </a> </li><li> <a href="/guide/kl-divergence" data-tier="2"> KL Divergence </a> </li><li> <a href="/guide/knowledge-graph" data-tier="3"> Knowledge Graph </a> </li><li> <a href="/guide/labeled-data" data-tier="2"> Labeled Data </a> </li><li> <a href="/guide/latent-variable" data-tier="3"> Latent Variable </a> </li><li> <a href="/guide/linear-regression" data-tier="2"> Linear Regression </a> </li><li> <a href="/guide/machine-learning" data-tier="1"> Machine Learning </a> </li><li> <a href="/guide/manifold-hypothesis" data-tier="3"> Manifold Hypothesis </a> </li><li> <a href="/guide/mcmc" data-tier="3"> Markov Chain Monte Carlo </a> </li><li> <a href="/guide/map" data-tier="3"> Maximum A Posteriori </a> </li><li> <a href="/guide/mle" data-tier="2"> Maximum Likelihood Estimation </a> </li><li> <a href="/guide/mutual-information" data-tier="3"> Mutual Information </a> </li><li> <a href="/guide/no-free-lunch" data-tier="3"> No Free Lunch Theorem </a> </li><li> <a href="/guide/normalization" data-tier="2"> Normalization </a> </li><li> <a href="/guide/occams-razor" data-tier="3"> Occam's Razor </a> </li><li> <a href="/guide/one-hot-encoding" data-tier="3"> One-Hot Encoding </a> </li><li> <a href="/guide/pac-learning" data-tier="3"> PAC Learning </a> </li><li> <a href="/guide/personalization" data-tier="3"> Personalization </a> </li><li> <a href="/guide/posterior" data-tier="3"> Posterior Distribution </a> </li><li> <a href="/guide/pca" data-tier="2"> Principal Component Analysis </a> </li><li> <a href="/guide/prior" data-tier="3"> Prior Distribution </a> </li><li> <a href="/guide/protein-folding" data-tier="3"> Protein Folding </a> </li><li> <a href="/guide/rademacher-complexity" data-tier="3"> Rademacher Complexity </a> </li><li> <a href="/guide/random-forest" data-tier="2"> Random Forest </a> </li><li> <a href="/guide/recommender-system" data-tier="2"> Recommender System </a> </li><li> <a href="/guide/regression" data-tier="1"> Regression </a> </li><li> <a href="/guide/semi-supervised-learning" data-tier="2"> Semi-Supervised Learning </a> </li><li> <a href="/guide/standardization" data-tier="2"> Standardization </a> </li><li> <a href="/guide/supervised-learning" data-tier="1"> Supervised Learning </a> </li><li> <a href="/guide/svm" data-tier="2"> Support Vector Machine </a> </li><li> <a href="/guide/synthetic-data" data-tier="3"> Synthetic Data </a> </li><li> <a href="/guide/time-series-forecasting" data-tier="3">
38 Time Series Forecasting </a> </li><li> <a href="/guide/training-data" data-tier="1"> Training Data </a> </li><li> <a href="/guide/unsupervised-learning" data-tier="1"> Unsupervised Learning </a> </li><li> <a href="/guide/variational-inference" data-tier="3"> Variational Inference </a> </li><li> <a href="/guide/vc-dimension" data-tier="3"> VC Dimension </a> </li><li> <a href="/guide/wasserstein-distance" data-tier="3"> Wasserstein Distance </a> </li> </ul> </details> </section><section id="region-neural-nets" class="fg-region" style="--rc: #ff6b35" aria-labelledby="h-neural-nets"> <div class="flex flex-wrap items-baseline justify-between gap-3"> <h2 id="h-neural-nets" class="fg-display flex items-center gap-3 text-3xl text-foreground"> <span class="fg-dot" style="width: 0.85rem; height: 0.85rem"></span> Neural Networks </h2> <p class="fg-mono text-muted-foreground"> 71 terms · 9 landmarks 39</p> </div> <p class="mt-2 max-w-2xl text-muted-foreground">Layers, activations, backprop: the building blocks of deep learning.</p> <div class="mt-6 grid gap-3 sm:grid-cols-2 lg:grid-cols-4"> <a href="/guide/neural-network" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 35</span> <span class="fg-display text-lg text-foreground">Neural Network</span> <span class="line-clamp-2 text-sm text-muted-foreground">A model built from layers of simple units, each taking a weighted sum of its inputs and applying a nonlinearity; training adjusts the weights so the whole stack maps inputs to useful outputs.</span> </a><a href="/guide/cnn" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 14</span> <span class="fg-display text-lg text-foreground">Convolutional Neural Network</span> <span class="line-clamp-2 text-sm text-muted-foreground">A neural network that scans images with small learned filters, reusing the same weights at every position to build up from edges to whole objects.</span> </a><a href="/guide/softmax" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 8</span> <span class="fg-display text-lg text-foreground">Softmax</span> <span class="line-clamp-2 text-sm text-muted-foreground">A function that turns a list of scores (logits) into probabilities that are all positive and sum to 1; the standard output of classifiers and language models.</span> </a><a href="/guide/rnn" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 7</span> <span class="fg-display text-lg text-foreground">Recurrent Neural Network</span> <span class="line-clamp-2 text-sm text-muted-foreground">A neural network architecture with loops that allow information to persist, designed for sequential data like text and time series.</span> </a><a href="/guide/activation-function" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 6</span> <span class="fg-display text-lg text-foreground">Activation Function</span> <span class="line-clamp-2 text-sm text-muted-foreground">A non-linear function applied to neuron outputs that introduces non-linearity, enabling networks to learn complex patterns.</span> </a><a href="/guide/deep-learning" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 5</span> <span class="fg-display text-lg text-foreground">Deep Learning</span> <span class="line-clamp-2 text-sm text-muted-foreground">A subset of machine learning that uses neural networks with multiple layers (deep neural networks) to learn hierarchical representations of data.</span> </a><a href="/guide/parameter" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 3</span> <span class="fg-display text-lg text-foreground">Parameter</span> <span class="line-clamp-2 text-sm text-muted-foreground">A number inside a model, a weight or a bias, that training adjusts to reduce the loss. Together, a model's parameters are everything it has learned.</span> </a><a href="/guide/representation-learning" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 2</span> <span class="fg-display text-lg text-foreground">Representation Learning</span> <span class="line-clamp-2 text-sm text-muted-foreground">Letting a model learn its own internal description of the data, such as embeddings or layered visual features, instead of relying on features designed by hand.</span> </a><a href="/guide/dropout" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 1</span> <span class="fg-display text-lg text-foreground">Dropout</span> <span class="line-clamp-2 text-sm text-muted-foreground">A regularization technique that randomly switches off units during training so the network can't lean on any one of them, which reduces overfitting.</span> </a> </div> <details class="fg-all mt-6"> <summary>All 71 Neural Networks terms</summary> <ul class="fg-terms mt-4"> <li> <a href="/guide/activation-function" data-tier="1"> Activation Function </a> </li><li> <a href="/guide/autoencoder" data-tier="2"> Autoencoder </a> </li><li> <a href="/guide/backward-pass" data-tier="2"> Backward Pass </a> </li><li> <a href="/guide/batch-normalization" data-tier="2"> Batch Normalization </a> </li><li> <a href="/guide/bias-parameter" data-tier="3"> Bias </a> </li><li> <a href="/guide/boltzmann-machine" data-tier="3"> Boltzmann Machine </a> </li><li> <a href="/guide/bottleneck" data-tier="3"> Bottleneck </a> </li><li> <a href="/guide/capsnet" data-tier="3"> Capsule Network </a> </li><li> <a href="/guide/convolution" data-tier="2"> Convolution </a> </li><li> <a href="/guide/cnn" data-tier="1"> Convolutional Neural Network </a> </li><li> <a href="/guide/dbn" data-tier="3"> Deep Belief Network </a> </li><li> <a href="/guide/deep-learning" data-tier="1"> Deep Learning </a> </li><li> <a href="/guide/depthwise-separable-conv" data-tier="3"> Depthwise Separable Convolution </a> </li><li> <a href="/guide/dilated-convolution" data-tier="3"> Dilated Convolution </a> </li><li> <a href="/guide/dropout" data-tier="1"> Dropout </a> </li><li> <a href="/guide/esn" data-tier="3"> Echo State Network </a> </li><li> <a href="/guide/elu" data-tier="3"> ELU </a> </li><li> <a href="/guide/energy-based-model" data-tier="2">
39 Energy-Based Model </a> </li><li> <a href="/guide/feature-map" data-tier="2"> Feature Map </a> </li><li> <a href="/guide/feedforward-network" data-tier="2"> Feedforward Network </a> </li><li> <a href="/guide/filter" data-tier="3"> Filter </a> </li><li> <a href="/guide/forward-pass" data-tier="2"> Forward Pass </a> </li><li> <a href="/guide/gru" data-tier="2"> Gated Recurrent Unit </a> </li><li> <a href="/guide/gelu" data-tier="2"> GELU </a> </li><li> <a href="/guide/gat" data-tier="3"> Graph Attention Network </a> </li><li> <a href="/guide/gnn" data-tier="2"> Graph Neural Network </a> </li><li> <a href="/guide/group-normalization" data-tier="3"> Group Normalization </a> </li><li> <a href="/guide/he-initialization" data-tier="3"> He Initialization </a> </li><li> <a href="/guide/hidden-layer" data-tier="2"> Hidden Layer </a> </li><li> <a href="/guide/hopfield-network" data-tier="3"> Hopfield Network </a> </li><li> <a href="/guide/kernel" data-tier="2"> Kernel </a> </li><li> <a href="/guide/latent-space" data-tier="2"> Latent Space </a> </li><li> <a href="/guide/layer" data-tier="2"> Layer </a> </li><li> <a href="/guide/leaky-relu" data-tier="2"> Leaky ReLU </a> </li><li> <a href="/guide/lsm" data-tier="3"> Liquid State Machine </a> </li><li> <a href="/guide/lstm" data-tier="2"> Long Short-Term Memory </a> </li><li> <a href="/guide/maxout" data-tier="3"> Maxout </a> </li><li> <a href="/guide/message-passing" data-tier="3"> Message Passing </a> </li><li> <a href="/guide/mish" data-tier="3"> Mish </a> </li><li> <a href="/guide/moe" data-tier="2"> Mixture of Experts </a> </li><li> <a href="/guide/mlp" data-tier="2"> Multi-Layer Perceptron </a> </li><li> <a href="/guide/neural-network" data-tier="1"> Neural Network </a> </li><li> <a href="/guide/neural-ode" data-tier="3"> Neural ODE </a> </li><li> <a href="/guide/node-embedding" data-tier="3"> Node Embedding </a> </li><li> <a href="/guide/padding" data-tier="3"> Padding </a> </li><li> <a href="/guide/parameter" data-tier="1"> Parameter </a> </li><li> <a href="/guide/perceptron" data-tier="2"> Perceptron </a> </li><li> <a href="/guide/pooling" data-tier="2"> Pooling </a> </li><li> <a href="/guide/prelu" data-tier="3"> PReLU </a> </li><li> <a href="/guide/rbf-network" data-tier="3"> Radial Basis Function Network </a> </li><li> <a href="/guide/receptive-field" data-tier="3"> Receptive Field </a> </li><li> <a href="/guide/rnn" data-tier="1"> Recurrent Neural Network </a> </li><li> <a href="/guide/relu" data-tier="2"> ReLU </a> </li><li> <a href="/guide/representation-learning" data-tier="1"> Representation Learning </a> </li><li> <a href="/guide/reservoir-computing" data-tier="2"> Reservoir Computing </a> </li><li> <a href="/guide/residual-connection" data-tier="2"> Residual Connection </a> </li><li> <a href="/guide/rbm" data-tier="3"> Restricted Boltzmann Machine </a> </li><li> <a href="/guide/sigmoid" data-tier="2"> Sigmoid </a> </li><li> <a href="/guide/skip-connection" data-tier="3"> Skip Connection </a> </li><li> <a href="/guide/softmax" data-tier="1"> Softmax </a> </li><li> <a href="/guide/spectral-normalization" data-tier="3"> Spectral Normalization </a> </li><li> <a href="/guide/snn" data-tier="3"> Spiking Neural Network </a> </li><li> <a href="/guide/stride" data-tier="3"> Stride </a> </li><li> <a href="/guide/swish" data-tier="3"> Swish </a> </li><li> <a href="/guide/transposed-convolution" data-tier="3"> Transposed Convolution </a> </li><li> <a href="/guide/universal-approximation" data-tier="3"> Universal Approximation Theorem </a> </li><li> <a href="/guide/vae" data-tier="2"> Variational Autoencoder </a> </li><li> <a href="/guide/weight" data-tier="2"> Weight </a> </li><li> <a href="/guide/weight-initialization" data-tier="2"> Weight Initialization </a> </li><li> <a href="/guide/weight-normalization" data-tier="3"> Weight Normalization </a> </li><li> <a href="/guide/xavier-initialization" data-tier="3"> Xavier Initialization </a> </li> </ul> </details> </section><section id="region-training" class="fg-region" style="--rc: #ffc857" aria-labelledby="h-training"> <div class="flex flex-wrap items-baseline justify-between gap-3"> <h2 id="h-training" class="fg-display flex items-center gap-3 text-3xl text-foreground"> <span class="fg-dot" style="width: 0.85rem; height: 0.85rem"></span> Training </h2> <p class="fg-mono text-muted-foreground"> 80 terms · 11 landmarks 40</p> </div> <p class="mt-2 max-w-2xl text-muted-foreground">Losses, optimizers, regularization, and how models actually learn.</p> <div class="mt-6 grid gap-3 sm:grid-cols-2 lg:grid-cols-4"> <a href="/guide/training" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 17</span> <span class="fg-display text-lg text-foreground">Training</span> <span class="line-clamp-2 text-sm text-muted-foreground">The process of fitting a model to data by repeatedly measuring how wrong its outputs are and adjusting its parameters to reduce that error.</span> </a><a href="/guide/gradient-descent" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 12</span> <span class="fg-display text-lg text-foreground">Gradient Descent</span> <span class="line-clamp-2 text-sm text-muted-foreground">An optimization method that repeatedly moves a model's parameters a small step in the direction that most reduces the loss.</span> </a><a href="/guide/loss-function" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 12</span> <span class="fg-display text-lg text-foreground">Loss Function</span> <span class="line-clamp-2 text-sm text-muted-foreground">A function that scores how wrong a model's prediction is as a single number, which training then works to make as small as possible.</span> </a><a href="/guide/overfitting" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 9</span> <span class="fg-display text-lg text-foreground">Overfitting</span> <span class="line-clamp-2 text-sm text-muted-foreground">When a model fits its training data too closely, noise included, so it scores well on examples it has seen and poorly on new ones.</span> </a><a href="/guide/pre-training" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 7</span> <span class="fg-display text-lg text-foreground">Pre-training</span> <span class="line-clamp-2 text-sm text-muted-foreground">Training a model on a large dataset (often self-supervised) before fine-tuning on specific tasks, enabling transfer learning.</span> </a><a href="/guide/fine-tuning" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 6</span> <span class="fg-display text-lg text-foreground">Fine-Tuning</span> <span class="line-clamp-2 text-sm text-muted-foreground">The process of further training a pre-trained model on a specific dataset to adapt it for a particular task or domain.</span> </a><a href="/guide/backpropagation" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 5</span> <span class="fg-display text-lg text-foreground">Backpropagation</span> <span class="line-clamp-2 text-sm text-muted-foreground">The algorithm for computing gradients of the loss with respect to network weights, enabling training through gradient descent.</span> </a><a href="/guide/learning-rate" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 4</span> <span class="fg-display text-lg text-foreground">Learning Rate</span> <span class="line-clamp-2 text-sm text-muted-foreground">The hyperparameter that sets how big a step gradient descent takes on each update; too high makes training unstable, too low makes it crawl.</span> </a><a href="/guide/regularization" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 3</span> <span class="fg-display text-lg text-foreground">Regularization</span> <span class="line-clamp-2 text-sm text-muted-foreground">Any change to training meant to make a model generalize better rather than fit its training data more closely, such as weight penalties, dropout or early stopping.</span> </a><a href="/guide/hyperparameter" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 2</span> <span class="fg-display text-lg text-foreground">Hyperparameter</span> <span class="line-clamp-2 text-sm text-muted-foreground">A setting chosen before training, such as the learning rate or batch size, that c
40ontrols how a model learns but isn't itself learned from the data.</span> </a><a href="/guide/self-supervised-learning" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 2</span> <span class="fg-display text-lg text-foreground">Self-Supervised Learning</span> <span class="line-clamp-2 text-sm text-muted-foreground">Training on unlabeled data by making the data supply its own answers: hide or alter part of an input and have the model predict or match it.</span> </a> </div> <details class="fg-all mt-6"> <summary>All 80 Training terms</summary> <ul class="fg-terms mt-4"> <li> <a href="/guide/adagrad" data-tier="3"> AdaGrad </a> </li><li> <a href="/guide/adam" data-tier="2"> Adam Optimizer </a> </li><li> <a href="/guide/adamw" data-tier="2"> AdamW </a> </li><li> <a href="/guide/automl" data-tier="3"> AutoML </a> </li><li> <a href="/guide/backpropagation" data-tier="1"> Backpropagation </a> </li><li> <a href="/guide/batch-gd" data-tier="3"> Batch Gradient Descent </a> </li><li> <a href="/guide/batch-size" data-tier="2"> Batch Size </a> </li><li> <a href="/guide/catastrophic-forgetting" data-tier="2"> Catastrophic Forgetting </a> </li><li> <a href="/guide/continual-learning" data-tier="3"> Continual Learning </a> </li><li> <a href="/guide/contrastive-learning" data-tier="2"> Contrastive Learning </a> </li><li> <a href="/guide/contrastive-loss" data-tier="2"> Contrastive Loss </a> </li><li> <a href="/guide/cosine-annealing" data-tier="3"> Cosine Annealing </a> </li><li> <a href="/guide/cross-entropy" data-tier="2"> Cross-Entropy Loss </a> </li><li> <a href="/guide/ctc-loss" data-tier="3">
40 CTC Loss </a> </li><li> <a href="/guide/curriculum-learning" data-tier="3"> Curriculum Learning </a> </li><li> <a href="/guide/cutmix" data-tier="3"> CutMix </a> </li><li> <a href="/guide/cutout" data-tier="3"> Cutout </a> </li><li> <a href="/guide/cyclical-lr" data-tier="3"> Cyclical Learning Rate </a> </li><li> <a href="/guide/data-augmentation" data-tier="2"> Data Augmentation </a> </li><li> <a href="/guide/data-parallelism" data-tier="2"> Data Parallelism </a> </li><li> <a href="/guide/distillation-temperature" data-tier="3"> Distillation Temperature </a> </li><li> <a href="/guide/distributed-training" data-tier="2"> Distributed Training </a> </li><li> <a href="/guide/early-stopping" data-tier="2"> Early Stopping </a> </li><li> <a href="/guide/erm" data-tier="3"> Empirical Risk Minimization </a> </li><li> <a href="/guide/epoch" data-tier="2"> Epoch </a> </li><li> <a href="/guide/exploding-gradient" data-tier="3"> Exploding Gradient </a> </li><li> <a href="/guide/fine-tuning" data-tier="1"> Fine-Tuning </a> </li><li> <a href="/guide/focal-loss" data-tier="2"> Focal Loss </a> </li><li> <a href="/guide/gradient-accumulation" data-tier="3"> Gradient Accumulation </a> </li><li> <a href="/guide/gradient-checkpointing" data-tier="3"> Gradient Checkpointing </a> </li><li> <a href="/guide/gradient-clipping" data-tier="3"> Gradient Clipping </a> </li><li> <a href="/guide/gradient-descent" data-tier="1"> Gradient Descent </a> </li><li> <a href="/guide/hinge-loss" data-tier="2"> Hinge Loss </a> </li><li> <a href="/guide/huber-loss" data-tier="2"> Huber Loss </a> </li><li> <a href="/guide/hyperparameter" data-tier="1"> Hyperparameter </a> </li><li> <a href="/guide/hyperparameter-tuning" data-tier="2"> Hyperparameter Tuning </a> </li><li> <a href="/guide/knowledge-distillation" data-tier="2"> Knowledge Distillation </a> </li><li> <a href="/guide/l1-regularization" data-tier="3"> L1 Regularization </a> </li><li> <a href="/guide/l2-regularization" data-tier="2"> L2 Regularization </a> </li><li> <a href="/guide/label-smoothing" data-tier="3"> Label Smoothing </a> </li><li> <a href="/guide/lamb" data-tier="3"> LAMB Optimizer </a> </li><li> <a href="/guide/learning-rate" data-tier="1"> Learning Rate </a> </li><li> <a href="/guide/learning-rate-schedule" data-tier="2"> Learning Rate Schedule </a> </li><li> <a href="/guide/lookahead" data-tier="3"> Lookahead Optimizer </a> </li><li> <a href="/guide/lora" data-tier="2"> LoRA </a> </li><li> <a href="/guide/loss-function" data-tier="1"> Loss Function </a> </li><li> <a href="/guide/mse" data-tier="2"> Mean Squared Error </a> </li><li> <a href="/guide/meta-learning" data-tier="3"> Meta-Learning </a> </li><li> <a href="/guide/metric-learning" data-tier="2"> Metric Learning </a> </li><li> <a href="/guide/mini-batch-gd" data-tier="2"> Mini-Batch Gradient Descent </a> </li><li> <a href="/guide/mixed-precision" data-tier="2"> Mixed Precision Training </a> </li><li> <a href="/guide/mixup" data-tier="3"> Mixup </a> </li><li> <a href="/guide/model-parallelism" data-tier="2"> Model Parallelism </a> </li><li> <a href="/guide/momentum" data-tier="2"> Momentum </a> </li><li> <a href="/guide/multi-task-learning" data-tier="3"> Multi-Task Learning </a> </li><li> <a href="/guide/nesterov-momentum" data-tier="3"> Nesterov Momentum </a> </li><li> <a href="/guide/nas" data-tier="3"> Neural Architecture Search </a> </li><li> <a href="/guide/online-learning" data-tier="3"> Online Learning </a> </li><li> <a href="/guide/optimizer" data-tier="2"> Optimizer </a> </li><li> <a href="/guide/overfitting" data-tier="1"> Overfitting </a> </li><li> <a href="/guide/peft" data-tier="2"> Parameter-Efficient Fine-Tuning (PEFT) </a> </li><li> <a href="/guide/pipeline-parallelism" data-tier="3"> Pipeline Parallelism </a> </li><li> <a href="/guide/pre-training" data-tier="1"> Pre-training </a> </li><li> <a href="/guide/regularization" data-tier="1"> Regularization </a> </li><li> <a href="/guide/rmsprop" data-tier="3"> RMSprop </a> </li><li> <a href="/guide/self-supervised-learning" data-tier="1"> Self-Supervised Learning </a> </li><li> <a href="/guide/sharded-data-parallel" data-tier="3"> Sharded Data Parallelism </a> </li><li> <a href="/guide/step-decay" data-tier="3"> Step Decay </a> </li><li> <a href="/guide/sgd" data-tier="2"> Stochastic Gradient Descent </a> </li><li> <a href="/guide/student-model" data-tier="3"> Student Model </a> </li><li> <a href="/guide/teacher-model" data-tier="3"> Teacher Model </a> </li><li> <a href="/guide/tensor-parallelism" data-tier="3"> Tensor Parallelism </a> </li><li> <a href="/guide/training" data-tier="1"> Training </a> </li><li> <a href="/guide/transfer-learning" data-tier="2"> Transfer Learning </a> </li><li> <a href="/guide/triplet-loss" data-tier="3"> Triplet Loss </a> </li><li> <a href="/guide/underfitting" data-tier="2"> Underfitting </a> </li><li> <a href="/guide/vanishing-gradient" data-tier="2"> Vanishing Gradient </a> </li><li> <a href="/guide/warmup" data-tier="3"> Warmup </a> </li><li> <a href="/guide/weight-decay" data-tier="2"> Weight Decay </a> </li><li> <a href="/guide/zero" data-tier="3"> ZeRO </a> </li> </ul> </details> </section><section id="region-evaluation" class="fg-region" style="--rc: #9ad97a" aria-labelledby="h-evaluation"> <div class="flex flex-wrap items-baseline justify-between gap-3"> <h2 id="h-evaluation" class="fg-display flex items-center gap-3 text-3xl text-foreground"> <span class="fg-dot" style="width: 0.85rem; height: 0.85rem"></span> Evaluation </h2> <p class="fg-mono text-muted-foreground"> 45 terms · 7 landmarks 41</p> </div> <p class="mt-2 max-w-2xl text-muted-foreground">Metrics, benchmarks, and knowing whether a model is any good.</p> <div class="mt-6 grid gap-3 sm:grid-cols-2 lg:grid-cols-4"> <a href="/guide/recall" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 3</span> <span class="fg-display text-lg text-foreground">Recall</span> <span class="line-clamp-2 text-sm text-muted-foreground">The share of actual positives a model correctly identifies, TP / (TP + FN); also called sensitivity or the true positive rate.</span> </a><a href="/guide/false-positive" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 2</span> <span class="fg-display text-lg text-foreground">False Positive</span> <span class="line-clamp-2 text-sm text-muted-foreground">A case the model labels positive that is actually negative, such as a legitimate email sent to spam. Statisticians call it a Type I error.</span> </a><a href="/guide/precision" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 2</span> <span class="fg-display text-lg text-foreground">Precision</span> <span class="line-clamp-2 text-sm text-muted-foreground">Of everything a model flagged as positive, the fraction that really was positive: TP / (TP + FP). It answers "when the model says yes, how often is it right?"</span> </a><a href="/guide/test-set" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 2</span> <span class="fg-display text-lg text-foreground">Test Set</span> <span class="line-clamp-2 text-sm text-muted-foreground">Data locked away until the end of a project and used once to estimate how the finished model will perform on new, unseen examples.</span> </a><a href="/guide/validation-set" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 2</span> <span class="fg-display text-lg text-foreground">Validation Set</span> <span class="line-clamp-2 text-sm text-muted-foreground">Data held back from training and used to choose hyperparameters and checkpoints, so those choices are judged on examples the model never fit.</span> </a><a href="/guide/accuracy" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 1</span> <span class="fg-display text-lg text-foreground">Accuracy</span> <span class="line-clamp-2 text-sm text-muted-foreground">The share of predictions a classifier gets right. Simple and intuitive, but misleading when one class is much rarer than the others.</span> </a><a href="/guide/iou" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 1</span> <span class="fg-display text-lg text-foreground">Intersection over Union</span> <span class="line-clamp-2 text-sm text-muted-foreground">
41A score from 0 to 1 for how well a predicted region matches the true one: the area they share divided by the area they cover together.</span> </a> </div> <details class="fg-all mt-6"> <summary>All 45 Evaluation terms</summary> <ul class="fg-terms mt-4"> <li> <a href="/guide/ab-testing" data-tier="2"> A/B Testing </a> </li><li> <a href="/guide/accuracy" data-tier="1"> Accuracy </a> </li><li> <a href="/guide/attention-visualization" data-tier="3"> Attention Visualization </a> </li><li> <a href="/guide/auc" data-tier="2"> AUC </a> </li><li> <a href="/guide/average-precision" data-tier="2"> Average Precision </a> </li><li> <a href="/guide/baseline" data-tier="2"> Baseline Model </a> </li><li> <a href="/guide/benchmark" data-tier="2"> Benchmark </a> </li><li> <a href="/guide/benchmark-gaming" data-tier="3"> Benchmark Gaming </a> </li><li> <a href="/guide/bleu" data-tier="2"> BLEU Score </a> </li><li> <a href="/guide/calibration" data-tier="2"> Calibration </a> </li><li> <a href="/guide/cohens-kappa" data-tier="3"> Cohen's Kappa </a> </li><li> <a href="/guide/confusion-matrix" data-tier="2"> Confusion Matrix </a> </li><li> <a href="/guide/cross-validation" data-tier="2"> Cross-Validation </a> </li><li> <a href="/guide/data-leakage" data-tier="2"> Data Leakage </a> </li><li> <a href="/guide/dice-coefficient" data-tier="3"> Dice Coefficient </a> </li><li> <a href="/guide/distribution-shift" data-tier="2"> Distribution Shift </a> </li><li> <a href="/guide/f1-score" data-tier="2"> F1 Score </a> </li><li> <a href="/guide/false-negative" data-tier="2"> False Negative </a> </li><li> <a href="/guide/false-positive" data-tier="1"> False Positive </a> </li><li> <a href="/guide/false-positive-rate" data-tier="2"> False Positive Rate </a> </li><li> <a href="/guide/grad-cam" data-tier="3"> Grad-CAM </a> </li><li> <a href="/guide/iou" data-tier="1"> Intersection over Union </a> </li><li> <a href="/guide/lime" data-tier="2"> LIME </a> </li><li> <a href="/guide/log-loss" data-tier="2"> Log Loss </a> </li><li> <a href="/guide/mcc" data-tier="3"> Matthews Correlation Coefficient </a> </li><li> <a href="/guide/mae" data-tier="3"> Mean Absolute Error </a> </li><li> <a href="/guide/mean-average-precision" data-tier="2"> Mean Average Precision (mAP) </a> </li><li> <a href="/guide/ood" data-tier="2"> Out-of-Distribution </a> </li><li> <a href="/guide/perplexity" data-tier="2"> Perplexity </a> </li><li> <a href="/guide/precision" data-tier="1"> Precision </a> </li><li> <a href="/guide/pr-curve" data-tier="2"> Precision-Recall Curve </a> </li><li> <a href="/guide/r-squared" data-tier="3"> R-squared </a> </li><li> <a href="/guide/recall" data-tier="1"> Recall </a> </li><li> <a href="/guide/robustness" data-tier="2"> Robustness </a> </li><li> <a href="/guide/roc-curve" data-tier="2"> ROC Curve </a> </li><li> <a href="/guide/rouge" data-tier="3"> ROUGE Score </a> </li><li> <a href="/guide/saliency-map" data-tier="2"> Saliency Map </a> </li><li> <a href="/guide/shap" data-tier="2"> SHAP </a> </li><li> <a href="/guide/specificity" data-tier="2"> Specificity </a> </li><li> <a href="/guide/test-set" data-tier="1"> Test Set </a> </li><li> <a href="/guide/train-test-split" data-tier="2"> Train-Test Split </a> </li><li> <a href="/guide/true-negative" data-tier="2"> True Negative </a> </li><li> <a href="/guide/true-positive" data-tier="2"> True Positive </a> </li><li> <a href="/guide/type-i-ii-errors" data-tier="2"> Type I and Type II Errors </a> </li><li> <a href="/guide/validation-set" data-tier="1"> Validation Set </a> </li> </ul> </details> </section><section id="region-language" class="fg-region" style="--rc: #3de0c5" aria-labelledby="h-language"> <div class="flex flex-wrap items-baseline justify-between gap-3"> <h2 id="h-language" class="fg-display flex items-center gap-3 text-3xl text-foreground"> <span class="fg-dot" style="width: 0.85rem; height: 0.85rem"></span> Language & LLMs </h2> <p class="fg-mono text-muted-foreground"> 118 terms · 14 landmarks 42</p> </div> <p class="mt-2 max-w-2xl text-muted-foreground">Tokens, attention, transformers, prompting, and large language models.</p> <div class="mt-6 grid gap-3 sm:grid-cols-2 lg:grid-cols-4"> <a href="/guide/llm" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 18</span> <span class="fg-display text-lg text-foreground">Large Language Model</span> <span class="line-clamp-2 text-sm text-muted-foreground">A neural network, almost always a transformer, trained on vast amounts of text to predict the next token, which lets it write, answer, summarize and follow instructions.
42</span> </a><a href="/guide/embedding" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 16</span> <span class="fg-display text-lg text-foreground">Embedding</span> <span class="line-clamp-2 text-sm text-muted-foreground">A list of numbers (a vector) that represents a word, sentence, image or other item, learned so that similar items end up close together.</span> </a><a href="/guide/nlp" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 11</span> <span class="fg-display text-lg text-foreground">Natural Language Processing</span> <span class="line-clamp-2 text-sm text-muted-foreground">The field of AI that lets computers read, interpret, translate and generate human language, from spam filters and search to chatbots.</span> </a><a href="/guide/transformer" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 10</span> <span class="fg-display text-lg text-foreground">Transformer</span> <span class="line-clamp-2 text-sm text-muted-foreground">A neural network architecture, introduced in 2017, built from stacked self-attention and feed-forward layers; the basis of nearly every modern large language model.</span> </a><a href="/guide/language-modeling" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 8</span> <span class="fg-display text-lg text-foreground">Language Modeling</span> <span class="line-clamp-2 text-sm text-muted-foreground">Assigning probabilities to sequences of text, in practice by predicting each next token from the ones before it; the objective behind large language models.</span> </a><a href="/guide/tokenization" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 8</span> <span class="fg-display text-lg text-foreground">Tokenization</span> <span class="line-clamp-2 text-sm text-muted-foreground">Splitting text into tokens, usually subword pieces, and mapping each to an integer ID so a language model can process it.</span> </a><a href="/guide/attention-mechanism" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 7</span> <span class="fg-display text-lg text-foreground">Attention Mechanism</span> <span class="line-clamp-2 text-sm text-muted-foreground">A technique that lets a neural network weigh every part of its input when producing each output, focusing on the parts most relevant at that step.</span> </a><a href="/guide/token" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 7</span> <span class="fg-display text-lg text-foreground">Token</span> <span class="line-clamp-2 text-sm text-muted-foreground">The unit of text a language model reads and writes: a word, part of a word, or a character, drawn from the model's fixed vocabulary.</span> </a><a href="/guide/self-attention" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 6</span> <span class="fg-display text-lg text-foreground">Self-Attention</span> <span class="line-clamp-2 text-sm text-muted-foreground">An attention step in which every token in a sequence looks at every token in that same sequence and builds a new representation from a weighted mix of them.</span> </a><a href="/guide/prompt-engineering" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 4</span> <span class="fg-display text-lg text-foreground">Prompt Engineering</span> <span class="line-clamp-2 text-sm text-muted-foreground">Designing a model's input (instructions, context, examples and output format) to get better results without changing the model itself.</span> </a><a href="/guide/context-window" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 2</span> <span class="fg-display text-lg text-foreground">Context Window</span> <span class="line-clamp-2 text-sm text-muted-foreground">The maximum number of tokens a language model can take into account at once, counting both the input it reads and the output it writes. Also called context length.</span> </a><a href="/guide/hallucination" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 1</span> <span class="fg-display text-lg text-foreground">Hallucination</span> <span class="line-clamp-2 text-sm text-muted-foreground">When a language model produces fluent, confident output that is false or unsupported by its sources, such as invented facts, quotes or citations.</span> </a><a href="/guide/rag" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 1</span> <span class="fg-display text-lg text-foreground">Retrieval-Augmented Generation</span> <span class="line-clamp-2 text-sm text-muted-foreground">Fetching relevant documents at question time and adding them to an LLM's prompt, so answers can draw on current, checkable sources without retraining.</span> </a><a href="/guide/foundation-model" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 0</span> <span class="fg-display text-lg text-foreground">Foundation Model</span> <span class="line-clamp-2 text-sm text-muted-foreground">A large model pre-trained on broad data at scale that can be adapted to many downstream tasks, such as GPT-style LLMs, BERT or
42CLIP.</span> </a> </div> <details class="fg-all mt-6"> <summary>All 118 Language & LLMs terms</summary> <ul class="fg-terms mt-4"> <li> <a href="/guide/alignment-tax" data-tier="3"> Alignment Tax </a> </li><li> <a href="/guide/anaphora-resolution" data-tier="3"> Anaphora Resolution </a> </li><li> <a href="/guide/assistant-response" data-tier="3"> Assistant Response </a> </li><li> <a href="/guide/attention-head" data-tier="2"> Attention Head </a> </li><li> <a href="/guide/attention-paper" data-tier="2"> Attention Is All You Need </a> </li><li> <a href="/guide/attention-mask" data-tier="3"> Attention Mask </a> </li><li> <a href="/guide/attention-mechanism" data-tier="1"> Attention Mechanism </a> </li><li> <a href="/guide/attention-score" data-tier="3"> Attention Score </a> </li><li> <a href="/guide/autoregressive" data-tier="2"> Autoregressive Model </a> </li><li> <a href="/guide/bart" data-tier="3"> BART </a> </li><li> <a href="/guide/beam-search" data-tier="2"> Beam Search </a> </li><li> <a href="/guide/bert" data-tier="2"> BERT </a> </li><li> <a href="/guide/bidirectional-attention" data-tier="3"> Bidirectional Attention </a> </li><li> <a href="/guide/bos-token" data-tier="3"> BOS Token </a> </li><li> <a href="/guide/bpe" data-tier="3"> BPE </a> </li><li> <a href="/guide/causal-language-modeling" data-tier="2"> Causal Language Modeling </a> </li><li> <a href="/guide/causal-mask" data-tier="3"> Causal Mask </a> </li><li> <a href="/guide/chain-of-thought" data-tier="2"> Chain-of-Thought </a> </li><li> <a href="/guide/chatbot" data-tier="2"> Chatbot </a> </li><li> <a href="/guide/cloze-task" data-tier="3"> Cloze Task </a> </li><li> <a href="/guide/code-generation" data-tier="2"> Code Generation </a> </li><li> <a href="/guide/constituency-parsing" data-tier="3"> Constituency Parsing </a> </li><li> <a href="/guide/constitutional-ai" data-tier="2"> Constitutional AI </a> </li><li> <a href="/guide/context-window" data-tier="1"> Context Window </a> </li><li> <a href="/guide/conversation-history" data-tier="3"> Conversation History </a> </li><li> <a href="/guide/coreference-resolution" data-tier="3"> Coreference Resolution </a> </li><li> <a href="/guide/cosine-similarity" data-tier="2"> Cosine Similarity </a> </li><li> <a href="/guide/cross-attention" data-tier="2"> Cross-Attention </a> </li><li> <a href="/guide/decoder-only" data-tier="2"> Decoder-Only Model </a> </li><li> <a href="/guide/decoding-strategies" data-tier="2"> Decoding Strategies </a> </li><li> <a href="/guide/dependency-parsing" data-tier="3"> Dependency Parsing </a> </li><li> <a href="/guide/dialogue-state-tracking" data-tier="3"> Dialogue State Tracking </a> </li><li> <a href="/guide/embedding" data-tier="1"> Embedding </a> </li><li> <a href="/guide/emergent-abilities" data-tier="2"> Emergent Abilities </a> </li><li> <a href="/guide/encoder-decoder" data-tier="2"> Encoder-Decoder </a> </li><li> <a href="/guide/encoder-only" data-tier="3"> Encoder-Only Model </a> </li><li> <a href="/guide/entity-linking" data-tier="3"> Entity Linking </a> </li><li> <a href="/guide/eos-token" data-tier="3"> EOS Token </a> </li><li> <a href="/guide/few-shot" data-tier="2"> Few-Shot Learning </a> </li><li> <a href="/guide/flash-attention" data-tier="3"> Flash Attention </a> </li><li> <a href="/guide/foundation-model" data-tier="1"> Foundation Model </a> </li><li> <a href="/guide/gpt" data-tier="2"> GPT </a> </li><li> <a href="/guide/greedy-decoding" data-tier="2"> Greedy Decoding </a> </li><li> <a href="/guide/grounding" data-tier="2"> Grounding </a> </li><li> <a href="/guide/hallucination" data-tier="1"> Hallucination </a> </li><li> <a href="/guide/in-context-learning" data-tier="2"> In-Context Learning </a> </li><li> <a href="/guide/information-extraction" data-tier="2"> Information Extraction </a> </li><li> <a href="/guide/information-retrieval" data-tier="2"> Information Retrieval </a> </li><li> <a href="/guide/instruction-following" data-tier="3"> Instruction Following </a> </li><li> <a href="/guide/instruction-tuning" data-tier="2"> Instruction Tuning </a> </li><li> <a href="/guide/intent-recognition" data-tier="3"> Intent Recognition </a> </li><li> <a href="/guide/language-modeling" data-tier="1"> Language Modeling </a> </li><li> <a href="/guide/language-understanding" data-tier="2"> Language Understanding </a> </li><li> <a href="/guide/llm" data-tier="1"> Large Language Model </a> </li><li> <a href="/guide/lda" data-tier="3"> Latent Dirichlet Allocation </a> </li><li> <a href="/guide/lemmatization" data-tier="3"> Lemmatization </a> </li><li> <a href="/guide/length-penalty" data-tier="3"> Length Penalty </a> </li><li> <a href="/guide/machine-translation" data-tier="2"> Machine Translation </a> </li><li> <a href="/guide/masked-language-modeling" data-tier="2"> Masked Language Modeling </a> </li><li> <a href="/guide/masking" data-tier="2"> Masking </a> </li><li> <a href="/guide/multi-head-attention" data-tier="2"> Multi-Head Attention </a> </li><li> <a href="/guide/n-gram" data-tier="3"> N-gram </a> </li><li> <a href="/guide/ner" data-tier="2"> Named Entity Recognition </a> </li><li> <a href="/guide/nli" data-tier="3"> Natural Language Inference </a> </li><li> <a href="/guide/nlp" data-tier="1"> Natural Language Processing </a> </li><li> <a href="/guide/scaling-laws" data-tier="2"> Neural Scaling Laws </a> </li><li> <a href="/guide/next-token-prediction" data-tier="2"> Next-Token Prediction </a> </li><li> <a href="/guide/padding-token" data-tier="3"> Padding Token </a> </li><li> <a href="/guide/paraphrase-detection" data-tier="3"> Paraphrase Detection </a> </li><li> <a href="/guide/pos-tagging" data-tier="3"> Part-of-Speech Tagging </a> </li><li> <a href="/guide/positional-encoding" data-tier="2"> Positional Encoding </a> </li><li> <a href="/guide/prompt-engineering" data-tier="1"> Prompt Engineering </a> </li><li> <a href="/guide/prompt-template" data-tier="3"> Prompt Template </a> </li><li> <a href="/guide/prompt-tuning" data-tier="3"> Prompt Tuning </a> </li><li> <a href="/guide/query-key-value" data-tier="2"> Query-Key-Value </a> </li><li> <a href="/guide/question-answering" data-tier="2"> Question Answering </a> </li><li>
42 <a href="/guide/reading-comprehension" data-tier="3"> Reading Comprehension </a> </li><li> <a href="/guide/relation-extraction" data-tier="3"> Relation Extraction </a> </li><li> <a href="/guide/repetition-penalty" data-tier="3"> Repetition Penalty </a> </li><li> <a href="/guide/rag" data-tier="1"> Retrieval-Augmented Generation </a> </li><li> <a href="/guide/rig" data-tier="3"> Retrieval-Interleaved Generation </a> </li><li> <a href="/guide/rlhf" data-tier="2"> RLHF </a> </li><li> <a href="/guide/scaled-dot-product" data-tier="3"> Scaled Dot-Product Attention </a> </li><li> <a href="/guide/self-attention" data-tier="1"> Self-Attention </a> </li><li> <a href="/guide/srl" data-tier="3"> Semantic Role Labeling </a> </li><li> <a href="/guide/semantic-search" data-tier="2"> Semantic Search </a> </li><li> <a href="/guide/semantic-similarity" data-tier="2"> Semantic Similarity </a> </li><li> <a href="/guide/sentencepiece" data-tier="3"> SentencePiece </a> </li><li> <a href="/guide/sentiment-analysis" data-tier="2"> Sentiment Analysis </a> </li><li> <a href="/guide/seq2seq" data-tier="2"> Sequence-to-Sequence </a> </li><li> <a href="/guide/shot" data-tier="2"> Shot </a> </li><li> <a href="/guide/slot-filling" data-tier="3"> Slot Filling </a> </li><li> <a href="/guide/special-token" data-tier="3"> Special Token </a> </li><li> <a href="/guide/speech-recognition" data-tier="2"> Speech Recognition </a> </li><li> <a href="/guide/stemming" data-tier="3"> Stemming </a> </li><li> <a href="/guide/stop-words" data-tier="3"> Stop Words </a> </li><li> <a href="/guide/subword-tokenization" data-tier="2"> Subword Tokenization </a> </li><li> <a href="/guide/system-prompt" data-tier="2"> System Prompt </a> </li><li> <a href="/guide/t5" data-tier="2"> T5 </a> </li><li> <a href="/guide/temperature" data-tier="2"> Temperature </a> </li><li> <a href="/guide/text-classification" data-tier="2"> Text Classification </a> </li><li> <a href="/guide/text-generation" data-tier="2"> Text Generation </a> </li><li> <a href="/guide/text-summarization" data-tier="3"> Text Summarization </a> </li><li> <a href="/guide/textual-entailment" data-tier="3"> Textual Entailment </a> </li><li> <a href="/guide/tf-idf" data-tier="3"> TF-IDF </a> </li><li> <a href="/guide/token" data-tier="1"> Token </a> </li><li> <a href="/guide/tokenization" data-tier="1"> Tokenization </a> </li><li> <a href="/guide/top-k" data-tier="2"> Top-k Sampling </a> </li><li> <a href="/guide/top-p" data-tier="3"> Top-p Sampling </a> </li><li> <a href="/guide/topic-modeling" data-tier="3"> Topic Modeling </a> </li><li> <a href="/guide/transformer" data-tier="1"> Transformer </a> </li><li> <a href="/guide/user-prompt" data-tier="3"> User Prompt </a> </li><li> <a href="/guide/vector-database" data-tier="2"> Vector Database </a> </li><li> <a href="/guide/vocabulary-size" data-tier="3"> Vocabulary Size </a> </li><li> <a href="/guide/wsd" data-tier="3"> Word Sense Disambiguation </a> </li><li> <a href="/guide/word2vec" data-tier="2"> Word2Vec </a> </li><li> <a href="/guide/wordpiece" data-tier="3"> WordPiece </a> </li><li> <a href="/guide/zero-shot" data-tier="2"> Zero-Shot Learning </a> </li> </ul> </details> </section><section id="region-vision" class="fg-region" style="--rc: #5aa9ff" aria-labelledby="h-vision"> <div class="flex flex-wrap items-baseline justify-between gap-3"> <h2 id="h-vision" class="fg-display flex items-center gap-3 text-3xl text-foreground"> <span class="fg-dot" style="width: 0.85rem; height: 0.85rem"></span> Vision & Multimodal </h2> <p class="fg-mono text-muted-foreground"> 54 terms · 6 landmarks 43</p> </div> <p class="mt-2 max-w-2xl text-muted-foreground">Seeing, generating, and mixing images, video, and audio.</p> <div class="mt-6 grid gap-3 sm:grid-cols-2 lg:grid-cols-4"> <a href="/guide/computer-vision" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 16</span> <span class="fg-display text-lg text-foreground">Computer Vision</span> <span class="line-clamp-2 text-sm text-muted-foreground">The field of AI that gets computers to extract meaning from images and video: what is in them, where it is, and how it moves.</span> </a><a href="/guide/object-detection" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 8</span> <span class="fg-display text-lg text-foreground">Object Detection</span> <span class="line-clamp-2 text-sm text-muted-foreground">Finding every object of interest in an image and giving each a class label, a confidence score and a boun
43ding box.</span> </a><a href="/guide/image-classification" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 3</span> <span class="fg-display text-lg text-foreground">Image Classification</span> <span class="line-clamp-2 text-sm text-muted-foreground">Assigning one label from a fixed set of categories to a whole image, such as "cat" or "defective part"; a core computer vision task.</span> </a><a href="/guide/image-generation" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 3</span> <span class="fg-display text-lg text-foreground">Image Generation</span> <span class="line-clamp-2 text-sm text-muted-foreground">Creating new images, often from a text prompt, with generative models such as GANs, VAEs and diffusion models that learn to turn random noise into realistic pictures.</span> </a><a href="/guide/diffusion-model" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 2</span> <span class="fg-display text-lg text-foreground">Diffusion Model</span> <span class="line-clamp-2 text-sm text-muted-foreground">A generative model trained to remove noise a step at a time, so it can turn pure random noise into a new image, audio clip or other sample.</span> </a><a href="/guide/semantic-segmentation" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 2</span> <span class="fg-display text-lg text-foreground">Semantic Segmentation</span> <span class="line-clamp-2 text-sm text-muted-foreground">Labelling every pixel in an image with a class, such as road, car or sky, so the output is a map of what is where rather than a single tag.</span> </a> </div> <details class="fg-all mt-6"> <summary>All 54 Vision & Multimodal terms</summary> <ul class="fg-terms mt-4"> <li> <a href="/guide/3d-reconstruction" data-tier="3"> 3D Reconstruction </a> </li><li> <a href="/guide/anchor-box" data-tier="3"> Anchor Box </a> </li><li> <a href="/guide/audio-processing" data-tier="2"> Audio Processing </a> </li><li> <a href="/guide/autonomous-vehicles" data-tier="3"> Autonomous Vehicles </a> </li><li> <a href="/guide/bounding-box" data-tier="2"> Bounding Box </a> </li><li> <a href="/guide/center-crop" data-tier="3"> Center Crop </a> </li><li> <a href="/guide/clip" data-tier="2"> CLIP </a> </li><li> <a href="/guide/color-jittering" data-tier="3"> Color Jittering </a> </li><li> <a href="/guide/computer-vision" data-tier="1"> Computer Vision </a> </li><li> <a href="/guide/depth-estimation" data-tier="3"> Depth Estimation </a> </li><li> <a href="/guide/diffusion-model" data-tier="1"> Diffusion Model </a> </li><li> <a href="/guide/discriminator" data-tier="3"> Discriminator </a> </li><li> <a href="/guide/efficientnet" data-tier="3"> EfficientNet </a> </li><li> <a href="/guide/face-detection" data-tier="3"> Face Detection </a> </li><li> <a href="/guide/facial-recognition" data-tier="3"> Facial Recognition </a> </li><li> <a href="/guide/faster-r-cnn" data-tier="3"> Faster R-CNN </a> </li><li> <a href="/guide/fpn" data-tier="3"> Feature Pyramid Network </a> </li><li> <a href="/guide/gan" data-tier="2"> Generative Adversarial Network </a> </li><li> <a href="/guide/generator" data-tier="3"> Generator </a> </li><li> <a href="/guide/image-augmentation" data-tier="2"> Image Augmentation </a> </li><li> <a href="/guide/image-classification" data-tier="1"> Image Classification </a> </li><li> <a href="/guide/image-generation" data-tier="1"> Image Generation </a> </li><li> <a href="/guide/image-inpainting" data-tier="3"> Image Inpainting </a> </li><li> <a href="/guide/image-normalization" data-tier="3"> Image Normalization </a> </li><li> <a href="/guide/image-preprocessing" data-tier="2"> Image Preprocessing </a> </li><li> <a href="/guide/imagenet" data-tier="2"> ImageNet </a> </li><li> <a href="/guide/inception" data-tier="3"> Inception </a> </li><li> <a href="/guide/instance-segmentation" data-tier="2"> Instance Segmentation </a> </li><li> <a href="/guide/keypoint-detection" data-tier="2"> Keypoint Detection </a> </li><li> <a href="/guide/latent-diffusion" data-tier="2"> Latent Diffusion </a> </li><li> <a href="/guide/mask-r-cnn" data-tier="3"> Mask R-CNN </a> </li><li> <a href="/guide/medical-diagnosis" data-tier="3"> Medical Diagnosis </a> </li><li> <a href="/guide/multimodal-model" data-tier="2"> Multimodal Model </a> </li><li> <a href="/guide/nms" data-tier="3"> Non-Maximum Suppression </a> </li><li> <a href="/guide/object-detection" data-tier="1"> Object Detection </a> </li><li> <a href="/guide/ocr" data-tier="3"> Optical Character Recognition </a> </li><li> <a href="/guide/optical-flow" data-tier="3"> Optical Flow </a> </li><li> <a href="/guide/pose-estimation" data-tier="3"> Pose Estimation </a> </li><li> <a href="/guide/r-cnn" data-tier="3"> R-CNN </a> </li><li> <a href="/guide/random-crop" data-tier="3"> Random Crop </a> </li><li> <a href="/guide/rpn" data-tier="3"> Region Proposal Network </a> </li><li> <a href="/guide/resnet" data-tier="2"> ResNet </a> </li><li> <a href="/guide/retinanet" data-tier="3"> RetinaNet </a> </li><li> <a href="/guide/semantic-segmentation" data-tier="1"> Semantic Segmentation </a> </li><li> <a href="/guide/squeeze-excitation" data-tier="3"> Squeeze-and-Excitation </a> </li><li> <a href="/guide/stable-diffusion" data-tier="2"> Stable Diffusion </a> </li><li> <a href="/guide/style-transfer" data-tier="3"> Style Transfer </a> </li><li> <a href="/guide/super-resolution" data-tier="3"> Super-Resolution </a> </li><li> <a href="/guide/text-to-speech" data-tier="3"> Text-to-Speech </a> </li><li> <a href="/guide/transfer-learning-vision" data-tier="2"> Transfer Learning in Vision </a> </li><li> <a href="/guide/u-net" data-tier="2"> U-Net </a> </li><li> <a href="/guide/vgg" data-tier="3"> VGG </a> </li><li> <a href="/guide/vit" data-tier="2"> Vision Transformer </a> </li><li> <a href="/guide/yolo" data-tier="2"> YOLO </a> </li> </ul> </details> </section><section id="region-agents" class="fg-region" style="--rc: #7b6cff" aria-labelledby="h-agents"> <div class="flex flex-wrap items-baseline justify-between gap-3"> <h2 id="h-agents" class="fg-display flex items-center gap-3 text-3xl text-foreground"> <span class="fg-dot" style="width: 0.85rem; height: 0.85rem"></span> Agents & RL </h2> <p class="fg-mono text-muted-foreground"> 32 terms · 6 landmarks 44</p> </div> <p class="mt-2 max-w-2xl text-muted-foreground">Reinforcement learning, planning, tool use, and agents that act.</p> <div class="mt-6 grid gap-3 sm:grid-cols-2 lg:grid-cols-4"> <a href="/guide/reinforcement-learning" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 16</span> <span class="fg-display text-lg text-foreground">Reinforcement Learning</span> <span class="line-clamp-2 text-sm text-muted-foreground">Learning through interaction with an environment, receiving rewards or penalties to learn optimal behavior policies.</span> </a><a href="/guide/policy" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 4</span> <span class="fg-display text-lg text-foreground">Policy</span> <span class="line-clamp-2 text-sm text-muted-foreground">The rule an RL agent uses to choose actions: a mapping from each state to an action, or to a probability distribution over actions.</span> </a><a href="/guide/reward" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 4</span> <span class="fg-display text-lg text-foreground">Reward</span> <span class="line-clamp-2 text-sm text-muted-foreground">The single number an environment sends back after each action, telling a reinforcement learning agent how good that step was; the agent learns to maximize its long-run total.</span> </a><a href="/guide/agent" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 3</span> <span class="fg-display text-lg text-foreground">Agent</span> <span class="line-clamp-2 text-sm text-muted-foreground">In reinforcement learning, the learner and decision-maker that observes an environment, takes actions and adjusts its behaviour to maximize cumulative reward.</span> </a><a href="/guide/tool-use" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 2</span> <span class="fg-display text-lg text-foreground">Tool Use</span> <span class="line-clamp-2 text-sm text-muted-foreground">A language model asking the application around it to run a function, such as search, a calculator or an API, then using the result in its answer.</sp
44an> </a><a href="/guide/ai-agent" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 0</span> <span class="fg-display text-lg text-foreground">AI Agent</span> <span class="line-clamp-2 text-sm text-muted-foreground">A system where a large language model decides its own next steps in a loop: calling tools, reading the results, and continuing until the task is done.</span> </a> </div> <details class="fg-all mt-6"> <summary>All 32 Agents & RL terms</summary> <ul class="fg-terms mt-4"> <li> <a href="/guide/actor-critic" data-tier="2"> Actor-Critic </a> </li><li> <a href="/guide/agent" data-tier="1"> Agent </a> </li><li> <a href="/guide/ai-agent" data-tier="1"> AI Agent </a> </li><li> <a href="/guide/alphago" data-tier="3"> AlphaGo </a> </li><li> <a href="/guide/behavioral-cloning" data-tier="3"> Behavioral Cloning </a> </li><li> <a href="/guide/bellman-equation" data-tier="2"> Bellman Equation </a> </li><li> <a href="/guide/dqn" data-tier="2"> Deep Q-Network </a> </li><li> <a href="/guide/domain-randomization" data-tier="3"> Domain Randomization </a> </li><li> <a href="/guide/environment" data-tier="2"> Environment </a> </li><li> <a href="/guide/exploration-exploitation" data-tier="2"> Exploration vs Exploitation </a> </li><li> <a href="/guide/function-calling" data-tier="2"> Function Calling </a> </li><li> <a href="/guide/hierarchical-rl" data-tier="3"> Hierarchical RL </a> </li><li> <a href="/guide/imitation-learning" data-tier="2"> Imitation Learning </a> </li><li> <a href="/guide/inverse-rl" data-tier="3"> Inverse Reinforcement Learning </a> </li><li> <a href="/guide/mdp" data-tier="2"> Markov Decision Process </a> </li><li> <a href="/guide/meta-rl" data-tier="3"> Meta-RL </a> </li><li> <a href="/guide/model-based-rl" data-tier="2"> Model-Based RL </a> </li><li> <a href="/guide/model-free-rl" data-tier="2"> Model-Free RL </a> </li><li> <a href="/guide/mcts" data-tier="3"> Monte Carlo Tree Search </a> </li><li> <a href="/guide/multi-agent-rl" data-tier="3"> Multi-Agent RL </a> </li><li> <a href="/guide/multi-armed-bandit" data-tier="2"> Multi-Armed Bandit </a> </li><li> <a href="/guide/offline-rl" data-tier="3"> Offline RL </a> </li><li> <a href="/guide/policy" data-tier="1"> Policy </a> </li><li> <a href="/guide/policy-gradient" data-tier="2"> Policy Gradient </a> </li><li> <a href="/guide/ppo" data-tier="2"> PPO </a> </li><li> <a href="/guide/q-learning" data-tier="2"> Q-Learning </a> </li><li> <a href="/guide/reinforcement-learning" data-tier="1"> Reinforcement Learning </a> </li><li> <a href="/guide/reward" data-tier="1"> Reward </a> </li><li> <a href="/guide/sim-to-real" data-tier="3"> Sim-to-Real Transfer </a> </li><li> <a href="/guide/tool-use" data-tier="1"> Tool Use </a> </li><li> <a href="/guide/value-function" data-tier="2"> Value Function </a> </li><li> <a href="/guide/world-model" data-tier="3"> World Model </a> </li> </ul> </details> </section><section id="region-shipping" class="fg-region" style="--rc: #ff5d8f" aria-labelledby="h-shipping"> <div class="flex flex-wrap items-baseline justify-between gap-3"> <h2 id="h-shipping" class="fg-display flex items-center gap-3 text-3xl text-foreground"> <span class="fg-dot" style="width: 0.85rem; height: 0.85rem"></span> Shipping AI </h2> <p class="fg-mono text-muted-foreground"> 71 terms · 1 landmarks 45</p> </div> <p class="mt-2 max-w-2xl text-muted-foreground">Serving, infrastructure, efficiency, safety, and running AI for real.</p> <div class="mt-6 grid gap-3 sm:grid-cols-2 lg:grid-cols-4"> <a href="/guide/inference" class="fg-landmark"> <span class="fg-mono text-muted-foreground">Landmark · leads to 12</span> <span class="fg-display text-lg text-foreground">Inference</span> <span class="line-clamp-2 text-sm text-muted-foreground">
45Running a trained model on new inputs to get predictions, with its weights frozen: the stage of a model's life that users actually interact with.</span> </a> </div> <details class="fg-all mt-6"> <summary>All 71 Shipping AI terms</summary> <ul class="fg-terms mt-4"> <li> <a href="/guide/adversarial-attack" data-tier="2"> Adversarial Attack </a> </li><li> <a href="/guide/adversarial-example" data-tier="3"> Adversarial Example </a> </li><li> <a href="/guide/adversarial-perturbation" data-tier="2"> Adversarial Perturbation </a> </li><li> <a href="/guide/adversarial-training" data-tier="2"> Adversarial Training </a> </li><li> <a href="/guide/ai-alignment" data-tier="2"> AI Alignment </a> </li><li> <a href="/guide/ai-ethics" data-tier="2"> AI Ethics </a> </li><li> <a href="/guide/ai-governance" data-tier="2"> AI Governance </a> </li><li> <a href="/guide/ai-safety" data-tier="2"> AI Safety </a> </li><li> <a href="/guide/algorithmic-accountability" data-tier="3"> Algorithmic Accountability </a> </li><li> <a href="/guide/backdoor-attack" data-tier="3"> Backdoor Attack </a> </li><li> <a href="/guide/batch-processing" data-tier="3"> Batch Processing </a> </li><li> <a href="/guide/bias-in-ai" data-tier="2"> Bias in AI </a> </li><li> <a href="/guide/black-box" data-tier="2"> Black Box </a> </li><li> <a href="/guide/blue-green-deployment" data-tier="3"> Blue-Green Deployment </a> </li><li> <a href="/guide/canary-deployment" data-tier="3"> Canary Deployment </a> </li><li> <a href="/guide/certified-robustness" data-tier="3"> Certified Robustness </a> </li><li> <a href="/guide/ci-cd-ml" data-tier="3"> CI/CD for ML </a> </li><li> <a href="/guide/compute" data-tier="2"> Compute </a> </li><li> <a href="/guide/concept-drift" data-tier="2"> Concept Drift </a> </li><li> <a href="/guide/content-moderation" data-tier="3"> Content Moderation </a> </li><li> <a href="/guide/data-drift" data-tier="2"> Data Drift </a> </li><li> <a href="/guide/data-pipeline" data-tier="2"> Data Pipeline </a> </li><li> <a href="/guide/data-poisoning" data-tier="2"> Data Poisoning </a> </li><li> <a href="/guide/data-versioning" data-tier="2"> Data Versioning </a> </li><li> <a href="/guide/differential-privacy" data-tier="3"> Differential Privacy </a> </li><li> <a href="/guide/edge-deployment" data-tier="2"> Edge Deployment </a> </li><li> <a href="/guide/experiment-tracking" data-tier="2"> Experiment Tracking </a> </li><li> <a href="/guide/explainability" data-tier="2"> Explainability </a> </li><li> <a href="/guide/fairness" data-tier="2"> Fairness </a> </li><li> <a href="/guide/feature-store" data-tier="3"> Feature Store </a> </li><li> <a href="/guide/federated-learning" data-tier="2"> Federated Learning </a> </li><li> <a href="/guide/flops" data-tier="3"> FLOPS </a> </li><li> <a href="/guide/gpu" data-tier="2"> GPU </a> </li><li> <a href="/guide/grpc" data-tier="3"> gRPC </a> </li><li> <a href="/guide/homomorphic-encryption" data-tier="3"> Homomorphic Encryption </a> </li><li> <a href="/guide/inference" data-tier="1"> Inference </a> </li><li> <a href="/guide/inference-latency" data-tier="2"> Inference Latency </a> </li><li> <a href="/guide/interpretability" data-tier="2"> Interpretability </a> </li><li> <a href="/guide/membership-inference" data-tier="3"> Membership Inference </a> </li><li> <a href="/guide/ml-security" data-tier="2"> ML Security </a> </li><li> <a href="/guide/mlops" data-tier="2"> MLOps </a> </li><li> <a href="/guide/model-caching" data-tier="3"> Model Caching </a> </li><li> <a href="/guide/model-card" data-tier="3"> Model Card </a> </li><li> <a href="/guide/model-compression" data-tier="2"> Model Compression </a> </li><li> <a href="/guide/deployment" data-tier="2"> Model Deployment </a> </li><li> <a href="/guide/model-drift" data-tier="3"> Model Drift </a> </li><li> <a href="/guide/model-endpoint" data-tier="2"> Model Endpoint </a> </li><li> <a href="/guide/model-extraction" data-tier="3"> Model Extraction </a> </li><li> <a href="/guide/model-inversion" data-tier="3"> Model Inversion </a> </li><li> <a href="/guide/model-lineage" data-tier="3"> Model Lineage </a> </li><li> <a href="/guide/model-monitoring" data-tier="2"> Model Monitoring </a> </li><li> <a href="/guide/performance-degradation" data-tier="2"> Model Performance Degradation </a> </li><li> <a href="/guide/model-registry" data-tier="2"> Model Registry </a> </li><li> <a href="/guide/model-reproducibility" data-tier="2"> Model Reproducibility </a> </li><li> <a href="/guide/model-retraining" data-tier="2"> Model Retraining </a> </li><li> <a href="/guide/model-serving" data-tier="2"> Model Serving </a> </li><li> <a href="/guide/model-versioning" data-tier="2"> Model Versioning </a> </li><li> <a href="/guide/model-watermarking" data-tier="3"> Model Watermarking </a> </li><li> <a href="/guide/neuromorphic-computing" data-tier="3"> Neuromorphic Computing </a> </li><li> <a href="/guide/prediction-confidence" data-tier="2"> Prediction Confidence </a> </li><li> <a href="/guide/privacy-preserving-ml" data-tier="2"> Privacy-Preserving ML </a> </li><li> <a href="/guide/pruning" data-tier="2"> Pruning </a> </li><li> <a href="/guide/quantization" data-tier="2"> Quantization </a> </li><li> <a href="/guide/release-strategy" data-tier="2"> Release Strategy </a> </li><li> <a href="/guide/request-batching" data-tier="3"> Request Batching </a> </li><li> <a href="/guide/rest-api" data-tier="3"> REST API </a> </li><li> <a href="/guide/smpc" data-tier="3"> Secure Multi-Party Computation </a> </li><li> <a href="/guide/shadow-deployment" data-tier="3"> Shadow Deployment </a> </li><li> <a href="/guide/throughput" data-tier="3"> Throughput </a> </li><li> <a href="/guide/tpu" data-tier="3"> TPU </a> </li><li> <a href="/guide/tee" data-tier="3"> Trusted Execution Environment </a> </li> </ul> </details> </section> </div> </div> </main> <footer class="py-4" aria-label="Footer" role="contentinfo" id="footer"> <div class="mx-auto flex max-w-5xl flex-col items-center justify-center gap-y-2 px-4 sm:flex-row sm:justify-between"> <ul class="flex flex-wrap gap-2 justify-around w-full mb-4 md:justify-center md:mb-0 md:w-auto" role="list" aria-label="social links"> <li aria-label="GitHub" role="listitem"> <a href="https://github.com/farezv?ref=personal-website" target="_blank" class="duration-300 ease-in-out inline-flex items-center justify-center gap-2 whitespace-nowrap rounded-full font-medium transition-all disabled:pointer-events-none disabled:opacity-50 [&_svg]:pointer-events-none [&_svg:not([class*='size-'])]:size-4 shrink-0 [&_svg]:shrink-0 outline-none focus-visible:border-ring focus-visible:ring-ring/50 focus-visible:ring-[3px] aria-invalid:ring-destructive/20 dark:aria-invalid:ring-destructive/40 aria-invalid:border-destructive hover:bg-accent hover:text-accent-foreground dark:hover:bg-accent/50 group text-2xl size-14 lg:size-9 lg:text-base" aria-label="GitHub" title="GitHub" target="_blank" rel="noopener noreferrer"> <svg width="1em" height="1em" class="size-4" data-icon="lucide:github">
45 <symbol id="ai:lucide:github" viewBox="0 0 24 24"><g fill="none" stroke="currentColor" stroke-linecap="round" stroke-linejoin="round" stroke-width="2"><path d="M15 22v-4a4.8 4.8 0 0 0-1-3.5c3 0 6-2 6-5.5c.08-1.25-.27-2.48-1-3.5c.28-1.15.28-2.35 0-3.5c0 0-1 0-3 1.5c-2.64-.5-5.36-.5-8 0C6 2 5 2 5 2c-.3 1.15-.3 2.35 0 3.5A5.4 5.4 0 0 0 4 9c0 3.5 3 5.5 6 5.5c-.39.49-.68 1.05-.85 1.65S8.93 17.38 9 18v4"/><path d="M9 18c-4.51 2-5-2-7-2"/></g></symbol><use href="#ai:lucide:github"></use> </svg> </a> </li><li aria-label="Email" role="listitem"> <a href="/cdn-cgi/l/email-protection#deb8bfacbba49eacbba4a9bfacbbf0b0bbaa" target="_blank" class="duration-300 ease-in-out inline-flex items-center justify-center gap-2 whitespace-nowrap rounded-full font-medium transition-all disabled:pointer-events-none disabled:opacity-50 [&_svg]:pointer-events-none [&_svg:not([class*='size-'])]:size-4 shrink-0 [&_svg]:shrink-0 outline-none focus-visible:border-ring focus-visible:ring-ring/50 focus-visible:ring-[3px] aria-invalid:ring-destructive/20 dark:aria-invalid:ring-destructive/40 aria-invalid:border-destructive hover:bg-accent hover:text-accent-foreground dark:hover:bg-accent/50 group text-2xl size-14 lg:size-9 lg:text-base" aria-label="Email" title="Email" target="_blank" rel="noopener noreferrer"> <svg width="1em" height="1em" class="size-4" data-icon="lucide:mail"> <symbol id="ai:lucide:mail" viewBox="0 0 24 24"><g fill="none" stroke="currentColor" stroke-linecap="round" stroke-linejoin="round" stroke-width="2"><path d="m22 7l-8.991 5.727a2 2 0 0 1-2.009 0L2 7"/><rect width="20" height="16" x="2" y="4" rx="2"/></g></symbol><use href="#ai:lucide:mail"></use> </svg> </a> </li><li aria-label="LinkedIn" role="listitem"> <a href="https://www.linkedin.com/in/farezv?ref=personal-website" target="_blank" class="duration-300 ease-in-out inline-flex items-center justify-center gap-2 whitespace-nowrap rounded-full font-medium transition-all disabled:pointer-events-none disabled:opacity-50 [&_svg]:pointer-events-none [&_svg:not([class*='size-'])]:size-4 shrink-0 [&_svg]:shrink-0 outline-none focus-visible:border-ring focus-visible:ring-ring/50 focus-visible:ring-[3px] aria-invalid:ring-destructive/20 dark:aria-invalid:ring-destructive/40 aria-invalid:border-destructive hover:bg-accent hover:text-accent-foreground dark:hover:bg-accent/50 group text-2xl size-14 lg:size-9 lg:text-base" aria-label="LinkedIn" title="LinkedIn" target="_blank" rel="noopener noreferrer"> <svg width="1em" height="1em" class="size-4" data-icon="lucide:message-circle-question"> <symbol id="ai:lucide:message-circle-question" viewBox="0 0 24 24"><g fill="none" stroke="currentColor" stroke-linecap="round" stroke-linejoin="round" stroke-width="2"><path d="M2.992 16.342a2 2 0 0 1 .094 1.167l-1.065 3.29a1 1 0 0 0 1.236 1.168l3.413-.998a2 2 0 0 1 1.099.092a10 10 0 1 0-4.777-4.719"/><path d="M9.09 9a3 3 0 0 1 5.83 1c0 2-3 3-3 3m.08 4h.01"/></g></symbol><use href="#ai:lucide:message-circle-question"></use> </svg> </a> </li><li aria-label="Instagram" role="listitem"> <a href="https://www.instagram.com/farezv?ref=personal-website" target="_blank" class="duration-300 ease-in-out inline-flex items-center justify-center gap-2 whitespace-nowrap rounded-full font-medium transition-all disabled:pointer-events-none disabled:opacity-50 [&_svg]:pointer-events-none [&_svg:not([class*='size-'])]:size-4 shrink-0 [&_svg]:shrink-0 outline-none focus-visible:border-ring focus-visible:ring-ring/50 focus-visible:ring-[3px] aria-invalid:ring-destructive/20 dark:aria-invalid:ring-destructive/40 aria-invalid:border-destructive hover:bg-accent hover:text-accent-foreground dark:hover:bg-accent/50 group text-2xl size-14 lg:size-9 lg:text-base" aria-label="Instagram" title="Instagram" target="_blank" rel="noopener noreferrer"> <svg width="1em" height="1em" class="size-4" data-icon="lucide:instagram"> <symbol id="ai:lucide:instagram" viewBox="0 0 24 24"><g fill="none" stroke="currentColor" stroke-linecap="round" stroke-linejoin="round" stroke-width="2"><rect width="20" height="20" x="2" y="2" rx="5" ry="5"/><path d="M16 11.37A4 4 0 1 1 12.63 8A4 4 0 0 1 16 11.37m1.5-4.87h.01"/></g></symbol><use href="#ai:lucide:instagram"></use> </svg> </a> </li><li aria-label="RSS" role="listitem"> <a href="/rss.xml" target="_blank" class="duration-300 ease-in-out inline-flex items-center justify-center gap-2 whitespace-nowrap rounded-full font-medium transition-all disabled:pointer-events-none disabled:opacity-50 [&_svg]:pointer-events-none [&_svg:not([class*='size-'])]:size-4 shrink-0 [&_svg]:shrink-0 outline-none focus-visible:border-ring focus-visible:ring-ring/50 focus-visible:ring-[3px] aria-invalid:ring-destructive/20 dark:aria-invalid:ring-destructive/40 aria-invalid:border-destructive hover:bg-accent hover:text-accent-foreground dark:hover:bg-accent/50 group text-2xl size-14 lg:size-9 lg:text-base" aria-label="RSS" title="RSS" target="_blank" rel="noopener noreferrer"> <svg width="1em" height="1em" class="size-4" data-icon="lucide:rss">
45 <symbol id="ai:lucide:rss" viewBox="0 0 24 24"><g fill="none" stroke="currentColor" stroke-linecap="round" stroke-linejoin="round" stroke-width="2"><path d="M4 11a9 9 0 0 1 9 9M4 4a16 16 0 0 1 16 16"/><circle cx="5" cy="19" r="1"/></g></symbol><use href="#ai:lucide:rss"></use> </svg> </a> </li> </ul> <div class="flex flex-wrap items-center justify-center gap-x-2 text-center"> <span class="text-muted-foreground text-sm" aria-label="copyright"> 462023 - 2026 © All rights reserved. 47</span> <div data-orientation="vertical" role="none" data-slot="separator-root" class="bg-border shrink-0 data-[orientation=horizontal]:h-px data-[orientation=horizontal]:w-full data-[orientation=vertical]:h-full data-[orientation=vertical]:w-px hidden h-4! sm:block"></div> <p class="text-muted-foreground text-sm" aria-label="made with love"> 48Made with â¤ï¸ in Austin and Vancouver 49</p> </div> </div> </footer> </div>
vendor: 109 bytes, line 49
49 <script data-cfasync="false" src="/cdn-cgi/scripts/5c5dd728/cloudflare-static/email-decode.min.js"></script>
49<script type="module" src="https://static.cloudflareinsights.com/beacon.min.js/v31edd6df95cf4e85bb4c19e7a9bdbcba1788362987495" integrity="sha512-iIg7k2xntmwu6/uSb5tpc/hySgZc4eoL31yB29W6tJFo2akwjPWcEqnCEdJvGexCL0KEQwVYv5BlowfhVz26hg==" data-cf-beacon='{"version":"2024.11.0","token":"77a5a7c0975e4c08bcfa33ed5fe20bf6","r":1,"spa":2}' crossorigin="anonymous"></script>
49 50</body></html>
50<script type="module">const p={foundations:{id:"foundations",label:"Foundations",blurb:"Core ML ideas, classic algorithms, and the data they learn from.",color:"#d9c7a1"},"neural-nets":{id:"neural-nets",label:"Neural Networks",blurb:"Layers, activations, backprop: the building blocks of deep learning.",color:"#ff6b35"},training:{id:"training",label:"Training",blurb:"Losses, optimizers, regularization, and how models actually learn.",color:"#ffc857"},evaluation:{id:"evaluation",label:"Evaluation",blurb:"Metrics, benchmarks, and knowing whether a model is any good.",color:"#9ad97a"},language:{id:"language",label:"Language & LLMs",blurb:"Tokens, attention, transformers, prompting, and large language models.",color:"#3de0c5"},vision:{id:"vision",label:"Vision & Multimodal",blurb:"Seeing, generating, and mixing images, video, and audio.",color:"#5aa9ff"},agents:{id:"agents",label:"Agents & RL",blurb:"Reinforcement learning, planning, tool use, and agents that act.",color:"#7b6cff"},shipping:{id:"shipping",label:"Shipping AI",blurb:"Serving, infrastructure, efficiency, safety, and running AI for real.",color:"#ff5d8f"}},l=document.querySelector("[data-fg-search]"),a=l?.querySelector("[data-fg-input]"),r=l?.querySelector("[data-fg-results]");let c=null,d=null,i=-1;const g=()=>d??=fetch("/guide/search.json").then(e=>e.json()).then(e=>{c=e}).catch(()=>{d=null});function h(e,t){const n=Math.max(0,...(e.a??[]).map(s=>u(s.toLowerCase(),t)-5));return Math.max(u(e.t.toLowerCase(),t),n,b(e,t))}function b(e,t){return e.s.includes(t.replace(/\s+/g,"-"))?30:e.d.toLowerCase().includes(t)?10:0}function u(e,t){return e===t?100:e.startsWith(t)?80-e.length/100:e.split(/[\s-]/).some(n=>n.startsWith(t))?60:e.includes(t)?40:0}const o=e=>e.replace(/&/g,"&").replace(/</g,"<");function m(){if(!r||!a||!c)return;const e=a.value.trim().toLowerCase();if(!e){r.hidden=!0,a.setAttribute("aria-expanded","false");return}const t=c.map(n=>({row:n,s:h(n,e)})).filter(n=>n.s>0).map(n=>({...n,s:n.s+(n.row.l==="1"?6:n.row.l==="2"?2:0)})).sort((n,s)=>s.s-n.s||n.row.t.localeCompare(s.row.t)).slice(0,8);i=t.length?0:-1,r.innerHTML=t.length?t.map(({row:n},s)=>`<li role="presentation"><a role="option" id="fg-opt-${s}" aria-selected="${s===0}" href="/guide/${n.s}"><span class="fg-dot" style="--rc:${p[n.r].color}"></span><span class="fg-results__term">${o(n.t)}</span><span class="fg-results__meta">${n.n?`${n.n} ${n.n===1?"stop":"stops"} to get here`:"Start here"}</span><span class="fg-results__desc">${o(n.d)}</span></a></li>`).join(""):`<li class="px-3 py-3 text-sm" style="color:#8a92a3">No term matches â${o(a.value)}â.</li>`,r.hidden=!1,a.setAttribute("aria-expanded","true"),a.setAttribute("aria-activedescendant",t.length?"fg-opt-0":"")}function f(e){const t=r?Array.from(r.querySelectorAll('[role="option"]')):[];!t.length||!a||(i=(i+e+t.length)%t.length,t.forEach((n,s)=>n.setAttribute("aria-selected",String(s===i))),t[i].scrollIntoView({block:"nearest"}),a.setAttribute("aria-activedescendant",t[i].id))}a?.addEventListener("focus",()=>void g());a?.addEventListener("input",()=>void g()?.then(m));a?.addEventListener("keydown",e=>{if(e.key==="ArrowDown")f(1);else if(e.key==="ArrowUp")f(-1);else if(e.key==="Enter"){const t=r?.querySelector('[aria-selected="true"]');t&&!r.hidden&&(location.href=t.href)}else if(e.key==="Escape"){r&&(r.hidden=!0);return}else return;e.preventDefault()});document.addEventListener("click",e=>{r&&l&&!l.contains(e.target)&&(r.hidden=!0)});document.addEventListener("keydown",e=>{const t=e.target;e.key==="/"&&!/INPUT|TEXTAREA|SELECT/.test(t.tagName)&&!t.isContentEditable&&(e.preventDefault(),a?.focus())});</script>
Line numbers count LF bytes from the start of the resource, as the search results do. Vendor segments are library code the classifier recognised; they are stored but not indexed. Bytes are shown as Latin1 characters, one per byte.