1<!DOCTYPE html> 2<html> 3<head> 4 <!-- Global site tag (gtag.js) - Google Analytics -->
vendor: 69 bytes, lines 4-5
4 5 <script async src="https://www.googletagmanager.com/gtag/js?id=
5UA-25091213-1
vendor: 17 bytes, lines 5-6
5"></script> 6
6<script> 7
vendor: 144 bytes, lines 7-11
7window.dataLayer = window.dataLayer || []; 8 function gtag(){dataLayer.push(arguments);} 9 gtag('js', new Date()); 10 11 gtag('config', '
11UA-25091213-1
vendor: 9 bytes, lines 11-12
11'); 12
12</script>
12 13 14 <meta charset="utf-8"> 15 <meta content="width=device-width, initial-scale=1" name="viewport"> 16 <link rel="shortcut icon" href="/favicon.ico" type="image/x-icon"> 17 <link rel="icon" href="/favicon.ico" type="image/x-icon"> 18 <link rel="stylesheet" href="/styles.css" type="text/css"> 19 <title>Ajay Jain</title> 20<!-- Global site tag (gtag.js) - Google Analytics -->
vendor: 64 bytes, lines 20-21
20 21<script async src="https://www.googletagmanager.com/gtag/js?id=
21UA-25091213-1
vendor: 12 bytes, line 21
21"></script>
22<script> 23
vendor: 134 bytes, lines 23-27
23window.dataLayer = window.dataLayer || []; 24 function gtag(){dataLayer.push(arguments);} 25 gtag('js', new Date()); 26 27 gtag('config', '
27UA-25091213-1
vendor: 4 bytes, line 27
27');
28</script>
28</head> 29<body> 30 <header> 31 <h1> 32 Ajay Jain 33 </h1> 34 <hr> 35 </header> 36 37 <ul class="links"> 38 <li><a href="https://twitter.com/ajayj_">Twitter</a></li> 39 <li> | </li> 40 <li><a href="https://scholar.google.com/citations?user=Ih7iLuUAAAAJ">Google Scholar</a></li> 41 <li> | </li> 42 <li><a href="https://github.com/ajayjain">GitHub</a></li> 43 <li> | </li> 44 <li><a href="https://www.linkedin.com/in/ajay-jain">LinkedIn</a></li> 45 <li> | </li> 46 <li><a href="cv.pdf">CV</a></li> 47 <li> | </li> 48 <li><a href="http://bit.ly/ajayj-portfolio">Fun things I made</a></li> 49 <li> | </li> 50 <li><a href="https://old.ajayjain.net">Older projects</a></li> 51 </ul> 52 53 54 <img src="https://d33wubrfki0l68.cloudfront.net/cfe7fff8bd56d1d8a8408a5b505ca13c4fc85a19/db8fd/images/ajay.jpg" class="photo" /> 55 56 <h3 class="role">Synthesizing visual worlds</h3> 57 <!-- <h3 class="role">Doctoral student (Ph.D.), UC Berkeley</h3> --> 58 <!-- <h3 class="role">NSF Graduate Research Fellow</h3> --> 59 <!-- <h3 class="role">Previous: NVIDIA Research, Google Brain, Uber ATG, Meta AI, MIT</h3> --> 60 61 <!-- p class="institute"> 62 Berkeley Artificial Intelligence Research (BAIR)<br> 63 Department of Electrical Engineering and Computer Science (EECS)<br> 64 UC Berkeley 65 </p --> 66 <!--p class="institute"> 67 Computer Science and Artificial Intelligence Laboratory (CSAIL)<br> 68 Department of Electrical Engineering and Computer Science (EECS)<br> 69 Massachusetts Institute of Technology 70 </p--> 71 72 <p class="bio">I'm an AI researcher at UC Berkeley, where I work on generative models. My work includes diffusion models, text-to-3D with NeRFs, and scalable ML systems. 73 <p class="bio">Previously, I was at Google Brain, NVIDIA Research, Uber ATG, and Facebook AI. I graduated from MIT with an S.B. in Computer Science and was a director of the nonprofit Machine Intelligence Community. At MIT, I did research at CSAIL and the Media Lab. My research is supported by the NSF Graduate Research Fellowship.</p> 74 <p class="email">Email: ajayj at berkeley dot edu</p> 75 76 <h3>News</h3> 77 <section class="news"> 78 <h5><span class="new">New!</span> <b>July 2022:</b> Dream Fields wins the Best Poster award at <a href="https://ai4cc.net/">AI4CC</a></h5> 79 <h5><span class="new">New!</span> <b>May 2022:</b> AdaCat accepted to UAI 2022</h5> 80 <h5><b>March 2022:</b> Dream Fields accepted to CVPR 2022</h5> 81 <h5><b>Aug 2021:</b> ContraCode accepted to EMNLP 2021</h5> 82 <h5><b>July 2021:</b> DietNeRF accepted to ICCV 2021</h5> 83 <!-- h5><span class="new">New!</span> <b>May 2021:</b> Will be interning at Google Brain</h5 --> 84 <!-- h5><b>September 2020:</b> Two papers on <a href="#DDPM">generative models</a> and <a href="#SGM">RL</a> accepted to NeurIPS 2020</h5 --> 85 <!-- <h5><b>May 2020:</b> <a href="#LMConv">LMConv paper</a> on image generation accepted to UAI 2020</h5> --> 86 <!-- <h5><b>January 2020:</b> <a href="#Checkmate">Checkmate paper</a> on memory-efficient DL accepted to MLSys 2020</h5> --> 87 <!-- <h5><b>September 2019:</b> <a href="#DRF-Net">DRF-Net paper</a> on behavior forecasting accepted to CoRL 2019</h5> --> 88 <!-- <h5><b>June 2019:</b> Presented a talk, <a href="#POCSNet">paper</a> at ISCA 2019 Workshop on <a href="https://mlforsystems.org">Machine Learning for Systems</a></h5> --> 89 <!-- <h5><b>June 2019:</b> Poster at ICML 2019 Workshop on <a href="https://deep-phenomena.org">Phenomena of Deep Learning</a></h5> --> 90 <!-- h5><b>June 2019:</b> Started my PhD at UC Berkeley</h5--> 91 <!-- <h5><b>Dec 2018:</b> <a href="#Revec">Revec</a> accepted to Compiler Construction 2019</h5> --> 92 <!-- h5><b>June 2018:</b> Headed to Toronto for a research internship at Uber Advanced Technologies Group</h5--> 93 <!-- h5><b>June 2017:</b> Joining Facebook's Applied Machine Learning team for the summer</h5--> 94 <!-- <h5><b>August 2016:</b> Started my undergrad at MIT</h5> --> 95 <!-- <h5><b>June 2016:</b> Interning at Juniper Networks this summer</h5> --> 96 </section> 97 98 <!--h3>Pre-prints</h3> 99 <section> 100 </section--> 101 102 <h3>Featured Publications <span class="note">* indicates equal contribution</span></h3> 103 <section> 104 <section class="citation featured"> 105 <div class="description"> 106 <h4>VectorFusion: Text-to-SVG by Abstracting Pixel-Based Diffusion Models</h4> 107 <h5><b>Ajay Jain</b>*, Amber Xie*, Pieter Abbeel</h5> 108 <h6><span class="conference arxiv">arXiv 2022</span></h6> 109 <p>Generate vector graphics (SVGs), pixel art and sketches from text using the pretrained Stable Diffusion model.</p> 110 <ul> 111 <li><a href="/vectorfusion/">[Website]</a></li> 112 <li><a href="/vectorfusion/gallery.html">[Gallery]</a></li> 113 <li>
113<a href="https://arxiv.org/abs/2211.11319">[Paper]</a></li> 114 </ul> 115 </div> 116 <a href="/vectorfusion/" class="citation-media"> 117 <video height="110" muted="" autoplay="" loop="" poster="https://pub-751dccf31fca4af7b5a452d19d49cf43.r2.dev/optimization/submarine.jpeg" playsinline=""> 118 <source src="https://pub-751dccf31fca4af7b5a452d19d49cf43.r2.dev/optimization/submarine_encode.mp4" type="video/mp4"> 119 Your browser does not support the video tag. 120 </video> 121 </a> 122 </section> 123 124 <section class="citation featured"> 125 <div class="description"> 126 <h4>DreamFusion: Text-to-3D using 2D Diffusion</h4> 127 <h5>Ben Poole, <b>Ajay Jain</b>, Jonathan T. Barron, Ben Mildenhall</h5> 128 <h6><span class="conference arxiv">arXiv 2022</span></h6> 129 <p>We optimize a NeRF from scratch using a pretrained text-to-image diffusion model to do text-to-3D generative modeling.</p> 130 <ul> 131 <li><a href="https://dreamfusion3d.github.io/">[Website]</a></li> 132 <li><a href="https://dreamfusion3d.github.io/gallery.html">[Gallery]</a></li> 133 <li><a href="https://arxiv.org/abs/2209.14988">[Paper]</a></li> 134 </ul> 135 </div> 136 <a href="https://dreamfusion3d.github.io/" class="citation-media"> 137 <video height="110" muted="" autoplay="" loop="" poster="https://dreamfusion-cdn.ajayj.com/dreamfusion_square_teaser.jpg" playsinline=""> 138 <source src="https://dreamfusion-cdn.ajayj.com/dreamfusion_square_teaser.mp4" type="video/mp4"> 139 Your browser does not support the video tag. 140 </video> 141 </a> 142 </section> 143 144 <section class="citation featured"> 145 <div class="description"> 146 <h4>Zero-Shot Text-Guided Object Generation with Dream Fields</h4> 147 <h5><b>Ajay Jain</b>, Ben Mildenhall, Jon Barron, Pieter Abbeel, Ben Poole</h5> 148 <h6><span class="conference cvpr">CVPR 2022</span> Conference on Computer Vision and Pattern Recognition</h6> 149 <p>We combine neural rendering with multi-modal image and text representations to synthesize diverse 3D objects solely from natural language descriptions.</p> 150 <ul> 151 <li><a href="/dreamfields/">[Website]</a></li> 152 <li><a href="https://arxiv.org/abs/2112.01455">[arXiv]</a></li> 153 <li><a href="https://www.youtube.com/watch?v=1Fke6w46tv4">[Overview video]</a></li> 154 <li><a href="https://github.com/google-research/google-research/tree/master/dreamfields">[Code]</a></li> 155 <li><a href="https://colab.research.google.com/drive/17GtPqdUCbG5CsmTnQFecPpoq_zpNKX7A?usp=sharing">[Colab demo]</a></li> 156 <li><a href="/dreamfields/assets/dreamfields_cvpr_poster.pdf">[Poster]</a></li> 157 </ul> 158 </div> 159 <a href="/dreamfields/" class="citation-media"> 160 <video height=110 muted="" autoplay="" loop="" class="bordered" alt="a sculpture of a rooster" poster="https://dreamfusion-cdn.ajayj.com/dreamfield_after.jpg"> 161 <source src="https://dreamfusion-cdn.ajayj.com/dreamfield_after.mp4" type="video/mp4"> 162 Your browser does not support the video tag. 163 </video> 164 </a> 165 <!-- <p><em>"a sculpture of a rooster"</em></p> --> 166 </section> 167 168 <section class="citation featured" id="DDPM"> 169 <div class="description"> 170 <h4>Denoising Diffusion Probabilistic Models</h4> 171 <h5>Jonathan Ho, <b>Ajay Jain</b>, Pieter Abbeel</h5> 172 <h6><span class="conference neurips">NeurIPS 2020</span> 34th Conference on Neural Information Processing Systems</h6> 173 <p>High-quality likelihood-based image generation; connect diffusion models to denoising score matching and Langevin dynamics; compression, reconstruction and interpolation</p> 174 <ul> 175 <li><a href="https://hojonathanho.github.io/diffusion/">[Website]</a></li> 176 <li><a href="https://arxiv.org/abs/2006.11239">[arXiv]</a></li> 177 <li><a href="https://slideslive.com/38936172">[Talk]</a></li> 178 <li><a href="images/diffusion_face_and_bedroom_video.mp4">[Sample video]</a></li> 179 <li><a href="https://github.com/hojonathanho/diffusion">
179[Code]</a></li> 180 </ul> 181 </div> 182 <a class="citation-media" href="images/diffusion_face_and_bedroom_video.mp4"> 183 <img src="https://d33wubrfki0l68.cloudfront.net/52717fb07257d1729a14abaeb44505067bab084a/cce70/images/diffusion_face_video_small_square.gif" alt="Sampling four 256x256 px photos of faces from a Denoising Diffusion Probabilistic Model." height=110> 184 </a> 185 </section> 186 187 </section> 188 189 <h3>More publications</h3> 190 <section> 191 <section class="citation"> 192 <div class="description"> 193 <h4>Journey to the BAOAB-limit: finding effective MCMC samplers for score-based models</h4> 194 <h5><b>Ajay Jain</b>*, Ben Poole*</h5> 195 <h6><span class="conference workshop">Score-based Methods @ NeurIPS 2022</span></h6> 196 <p>Sample diffusion models with a single noise level + high diversity.</p> 197 <ul> 198 <li><a href="/journey/">[Website]</a></li> 199 <li><a href="https://colab.research.google.com/drive/17kesyBVqubV_Zzchf2XoR-7MHk5jxTuo?usp=sharing">[Colab]</a></li> 200 <li><a href="https://github.com/ajayjain/journey-diffusion-samplers">[Code]</a></li> 201 <li><a href="/journey/Journey to the BAOAB-limit finding effective MCMC samplers for score-based models.pdf">[Paper]</a></li> 202 </ul> 203 </div> 204 <a href="/journey/" class="citation-media"> 205 <video height="110" muted="" autoplay="" loop="" poster="https://pub-14c9ec3d906f4413affd3ae6bba970a7.r2.dev/macarons.jpeg" playsinline=""> 206 <source src="https://pub-14c9ec3d906f4413affd3ae6bba970a7.r2.dev/macarons gs 3, 5k steps, baoab-limit, t300 encoded.mp4" type="video/mp4"> 207 Your browser does not support the video tag. 208 </video> 209 </a> 210 </section> 211 212 <section class="citation"> 213 <div class="description"> 214 <h4>Adaptive Categorical Discretization for Autoregressive Models</h4> 215 <h5>Colin Li, <b>Ajay Jain</b>, Pieter Abbeel</h5> 216 <h6><span class="conference uai">UAI 2022</span> Conference on Uncertainty in Artificial Intelligence</h6> 217 <p>AdaCat learns expressive autoregressive models by capturing fine-grained variation in continuous distributions with discrete density estimators.</p> 218 <ul> 219 <li><a href="https://openreview.net/forum?id=HMzzPOLs9l5">[Paper]</a></li> 220 <li><a href="https://colinqiyangli.github.io/adacat/">[Website]</a></li> 221 <li><a href="https://github.com/ColinQiyangLi/AdaCat">[Code]</a></li> 222 <li><a href="/adacat/adacat_uai_poster.pdf">[Poster]</a></li> 223 </ul> 224 </div> 225 <a href="https://colinqiyangli.github.io/adacat/" class="citation-media"> 226 <img src="https://d33wubrfki0l68.cloudfront.net/9c9b711dd48615853cfc84ab6b399a2a56b7f1b7/b31eb/adacat/demo.gif" alt="Optimizing a flexible 1D AdaCat distribution." width=110> 227 </a> 228 </section> 229 230 <section class="citation"> 231 <div class="description"> 232 <h4>Contrastive Code Representation Learning</h4> 233 <h5>Paras Jain*, <b>Ajay Jain</b>*, Tianjun Zhang, Pieter Abbeel, Joseph E. Gonzalez, Ion Stoica</h5> 234 <h6><span class="conference emnlp">EMNLP 2021</span> Empirical Methods in Natural Language Processing</h6> 235 <p>Learn to represent software functionality for automated software engineering tasks like type inference, clone detection and summarization. Improving robustness of ML4Code.</p> 236 <ul> 237 <li><a href="https://parasj.github.io/contracode/">[Website]</a></li> 238 <li><a href="https://aclanthology.org/2021.emnlp-main.482/">[ACL]</a></li> 239 <li><a href="https://arxiv.org/abs/2007.04973">[arXiv]</a></li> 240 <li><a href="https://ajayjain.github.io/assets/2021-contracode-poster-emnlp.pdf">[Poster]</a></li> 241 <li><a href="https://ajayjain.github.io/assets/2021.10.11-contracode-slides-emnlp.pdf">[Slides]</a></li> 242 <li><a href="https://www.youtube.com/watch?v=0deUDO6BqYY">[Talk]</a></li> 243 <li><a href="https://github.com/parasj/contracode/">[Code]</a></li> 244 </ul> 245 </div> 246 <a href="https://parasj.github.io/contracode/" class="citation-media"> 247 <img src="https://d33wubrfki0l68.cloudfront.net/541535e70398430a3025313fe619fb3e84420422/80d72/images/contracode.png" alt="Conceptual diagram illustrating Contrastive Code Representation Learning." width=110> 248 </a> 249 </section> 250 251 <section class="citation"> 252 <div class="description"> 253 <h4>Putting NeRF on a Diet: Semantically Consistent Few-Shot View Synthesis</h4> 254 <h5><b>Ajay Jain</b>, Matthew Tancik, Pieter Abbeel</h5> 255 <h6><span class="conference iccv">ICCV 2021</span> International Conference on Computer Vision</h6> 256 <p>CLIP + NeRF: Given only a few images of an object or scene, we reconstruct its 3D structure & render novel views using prior knowledge contained in large image encoders.</p> 257 <ul> 258 <li><a href="/dietnerf/">[Website]</a></li> 259 <li><a href="https://arxiv.org/abs/2104.00677">[arXiv]</a></li> 260 <li><a href="https://github.com/ajayjain/DietNeRF/">[Code]</a></li> 261 <li><a href="https://www.youtube.com/watch?v=RF_3hsNizqw">[Overview video]</a></li> 262 </ul> 263 </div> 264 <a href="/dietnerf/" class="citation-media"> 265 <img src="https://d33wubrfki0l68.cloudfront.net/8f2a5fbbb0ccbae01a0f5402cef18a682db8905a/1c039/dietnerf/assets/img/dietnerf_method_anim_25p.gif" alt="Overview of DietNeRF's semantic consistency loss." width=110> 266 </a> 267 </section> 268 269 <section class="citation" id="SGM"> 270 <div class="description"> 271 <h4>Sparse Graphical Memory for Robust Planning</h4> 272 <h5>Scott Emmons*, <b>Ajay Jain</b>*, Michael Laskin*, Thanard Kurutach, Pieter Abbeel, Deepak Pathak</h5> 273 <h6><span class="conference neurips">NeurIPS 2020</span> 34th Conference on Neural Information Processing Systems</h6> 274 <p>Provably robust+efficient long-horizon monocular navigation combining sparse graph planning and RL. Propose two-way consistency to find landmark memories and create a topological map.</p> 275 <ul> 276 <li><a href="https://mishalaskin.github.io/sgm/">[Website]</a></li> 277 <li><a href="https://arxiv.org/abs/2003.06417">[arXiv]</a></li> 278 <li><a href="https://papers.nips.cc/paper/2020/hash/385822e359afa26d52b5b286226f2cea-Abstract.html">[NeurIPS]</a></li> 279 <li><a href="https://slideslive.com/38936447">[Talk]</a></li> 280 <li><a href="https://youtu.be/n3F3i7F3Lcg">[Video]</a></li> 281 <li><a href="https://github.io/ajayjain/assets/2020-sgm-poster-neurips.pdf">[Poster]</a></li> 282 <li><a href="https://github.com/scottemmons/sgm">[Code]</a></li> 283 </ul> 284 </div> 285 <a href="https://mishalaskin.github.io/sgm/" class="citation-media"> 286 <img src="https://d33wubrfki0l68.cloudfront.net/b71e320294b25174b29a3d092cd2836848929075/dd313/images/sgm_cleanup.gif" alt="Cleaning up errors in the sparse graphical memory data structure." height=110> 287 </a> 288 </section> 289 290 <section class="citation" id="LMConv"> 291 <div class="description"> 292 <h4>Locally Masked Convolution for Autoregressive Models</h4> 293 <h5><b>Ajay Jain</b>, Pieter Abbeel, Deepak Pathak</h5> 294 <h6><span class="conference uai">UAI 2020</span> 36th Conference on Uncertainty in Artificial Intelligence</h6> 295 <p>Outpainting with PixelCNNs. Our efficent op allows PixelCNNs to generate images in arbitrary orders.</p> 296 <ul> 297 <li><a href="https://ajayjain.github.io/lmconv/">[Website]</a></li> 298 <li><a href="https://arxiv.org/abs/2006.12486">[arXiv]</a></li> 299 <li><a href="https://proceedings.mlr.press/v124/jain20b.html">[PMLR]</a></li> 300 <li><a href="https://www.youtube.com/watch?v=uGRRv3SExSs">[Talk]</a></li> 301 <li><a href="https://github.com/ajayjain/lmconv">[Code]</a></li> 302 </ul> 303 </div> 304 <a href="https://ajayjain.github.io/lmconv/" class="citation-media"> 305 <img src="https://d33wubrfki0l68.cloudfront.net/4061396a135d425092ec9d4f251914f0151bbae9/0e60a/images/lmconv.png" alt="Locally Masked Convolution applies customizable masks to patches of an image" width=110> 306 </a> 307 </section> 308 309 <section class="citation" id="Checkmate"> 310 <div class="description"> 311 <h4>Checkmate: Breaking the Memory Wall with Optimal Tensor Rematerialization</h4> 312 <h5>Paras Jain*, <b>Ajay Jain</b>*, Aniruddha Nrusimha, Amir Gholami, Pieter Abbeel, Kurt Keutzer, Ion Stoica, Joseph E. Gonzalez</h5> 313 <h6><span class="conference mlsys">MLSys 2020</span> 3rd Conference on Machine Learning and Systems</h6> 314 <p>Use up to 5x less memory when training DNNs by recomputing activations</p> 315 <ul> 316 <li><a href="https://checkmateai.github.io/">[Website]</a></li> 317 <li><a href="https://arxiv.org/abs/1910.02653">[arXiv]</a></li> 318 <li><a href="https://ajayjain.github.io/assets/2020-checkmate-slides-mlsys.pdf">[Slides]</a></li> 319 <li><a href="slides/checkmate_slides.pptx">[PPTX]</a></li> 320 <li><a href="https://github.com/parasj/checkmate">[Code]</a></li> 321 </ul> 322 </div> 323 <a href="https://checkmateai.github.io/" class="citation-media"> 324 <img src="https://d33wubrfki0l68.cloudfront.net/54d5e3a3619df537c9095ad229acfe1e06e2721e/50ed7/images/2_illustrated_example.svg" alt="" width=110> 325 </a> 326 </section> 327 328 <section class="citation" id="DRF-Net"> 329 <div class="description"> 330 <h4>Discrete Residual Flow for Probabilistic Pedestrian Behavior Prediction</h4> 331 <h5><b>Ajay Jain</b>*, Sergio Casas Romero*, Renjie Liao*, Yuwen Xiong*, Song Feng, Sean Segal, Raquel Urtasun</h5> 332 <h6><span class="conference corl">CoRL 2019</span> 3rd Conference on Robot Learning, <span style="color: red">Spotlight talk</span></h6> 333 <p>Multimodal, long-range behavior forecasts by predicting state marginals</p> 334 <ul> 335 <li><a href="https://arxiv.org/abs/1910.08041">[arXiv]</a></li> 336 <li><a href="posters/DRFNet_CoRL2019_poster.pdf">[Poster]</a></li> 337 <li><a href="https://youtu.be/n1ehwLhKOLo">[Talk]</a></li> 338 <li><a href="https://ajayjain.github.io/assets/2019-drfnet-slides-corl.pdf">[Slides]</a></li> 339 </ul> 340 </div> 341 <img class="citation-media" src="https://d33wubrfki0l68.cloudfront.net/d032ae9595340c8e0f67c26c1dec27b29cfda82e/f5971/images/drfnet_square.png" alt="DRF-Net 10 second forecast" width=110> 342 </section> 343 344 <section class="citation" id="Revec"> 345 <div class="description"> 346 <h4>Revec: Program Rejuvenation through Revectorization</h4> 347 <h5>Charith Mendis*, <b>Ajay Jain</b>*, Paras Jain, Saman Amarasinghe</h5> 348 <h6><span class="conference cc">CC 2019</span> 28th International Conference on Compiler Construction</h6> 349 <p>Achieve performance portability for hand-vectorized programs, with up to 1.88x speedup</p> 350 <ul> 351 <li><a href="https://www.nextgenvec.org">[Website]</a></li> 352 <li><a href="https://dl.acm.org/citation.cfm?id=3307357">[Paper]</a></li> 353 <li><a href="https://arxiv.org/abs/1902.02816">[arXiv]</a></li> 354 <li><a href="https://www.youtube.com/watch?v=EoVozNOnr4k">[Talk]</a></li> 355 <li><a href="slides/revec_cc19_slides.pdf">[Slides]</a></li> 356 <li><a href="https://github.com/revec">[Code]</a></li> 357 </ul> 358 </div> 359 <a href="https://www.nextgenvec.org" class="citation-media"><img src="https://d33wubrfki0l68.cloudfront.net/2b2c967fd48df681f264ed1cd971eaacad637c7b/9de39/images/revec.png" alt="" height=110></a> 360 </section> 361 362 <section class="citation"> 363 <div class="description"> 364 <h4>Autonomy for Surface Ship Interception</h4> 365 <h5>C. Mirabito, D.N. Subramani, T. Lolla, J. P.J. Haley, <b>A. Jain</b>, P.F.J. Lermusiaux, C. Li, D. Yue, Y. Liu, F. Hover, N. Pulsone, J. Edwards, K. Railey, and G. Shaw</h5> 366 <h6><span class="conference oceans">OCEANS 2017</span> 60th OCEANS Conference, MTS/IEEE Aberdeen</h6> 367 <p>Time-optimal path planning for underwater robots</p> 368 <ul> 369 <li><a href="https://ieeexplore.ieee.org/document/8084817">[Paper]</a></li> 370 <li><a href="http://mseas.mit.edu/publications/PDF/mirabito_etal_autonomy_intercept_Oceans2017.pdf">[Open access]</a></li> 371 </ul> 372 </div> 373 <img class="citation-media" src="https://d33wubrfki0l68.cloudfront.net/740f4ab1407eb9d570a34c8d4cd2e7bb3237284e/10523/images/paths.png" alt="" height=110> 374 </section> 375 </section> 376 377 <h3>Short papers</h3> 378 <section> 379 <section class="citation" id="POCSNet"> 380 <div class="description"> 381 <h4>Learning Automatic Schedulers with Projective Reparameterization</h4> 382 <h5><b>Ajay Jain</b>, Saman Amarasinghe</h5> 383 <h6><span class="conference workshop isca">ISCA 2019</span> 46th International Symposium on Computer Architecture</h6> 384 <h6>Workshop on Machine Learning for Systems, Jun 2019, Talk</h6> 385 <p>Supervised learning of schedulers, with correctness constraints</p> 386 <ul> 387 <li><a href="papers/projective_isca19.pdf">[Paper]</a></li> 388 <li><a href="slides/pocsnet_isca19_slides.pdf">[Slides]</a></li> 389 <li><a href="http://mlforsystems.org/assets/slides/isca2019/MLforSystems2019_Talk_Ajay_Jain.pptx">[PPTX]</a></li> 390 </ul> 391 </div> 392 <img class="citation-media" src="https://d33wubrfki0l68.cloudfront.net/5a60dc1ba36c5b616d64384c234811806ba95ceb/7e6dc/images/projection.png" alt="" height=110> 393 </section> 394 395 <section class="citation"> 396 <div class="description"> 397 <h4>Using effective dimension to analyze feature transformations in deep neural networks</h4> 398 <h5>Kavya Ravichandran, <b>Ajay Jain</b>, Alexander Rakhlin</h5> 399 <h6><span class="conference workshop icml">ICML 2019</span> 36th International Conference on Machine Learning</h6> 400 <h6>Workshop on Identifying and Understanding Deep Learning Phenomena, Jun 2019</h6> 401 <ul> 402 <li><a href="https://openreview.net/pdf?id=HJGsj13qTE">[Paper]</a></li> 403 </ul> 404 </div> 405 </section> 406 407 <section class="citation"> 408 <div class="description"> 409 <h4>The Case for GPU Multitenancy</h4> 410 <h5>Paras Jain, Xiangxi Mo, <b>Ajay Jain</b>, Alexey Tumanov, Joseph E. Gonzalez, Ion Stoica</h5> 411 <h6><span class="conference arxiv">arXiv 2019</span> arXiv:1910.02653, Jan 2019</h6> 412 <ul> 413 <li><a href="https://arxiv.org/abs/1901.10008">[arXiv]</a></li> 414 </ul> 415 </div> 416 <img class="citation-media" src="https://d33wubrfki0l68.cloudfront.net/3c8417738473f734cdd1ff8c994da4cf9fc6c9a3/a9951/images/fijit.png" alt="" height=100> 417 </section> 418 419 <secti
419on class="citation"> 420 <div class="description"> 421 <h4>Dynamic Space-Time scheduling for GPU inference</h4> 422 <h5>Paras Jain, Xiangxi Mo, <b>Ajay Jain</b>, Harikaran Subbaraj, Rehan Sohail Durrani, Alexey Tumanov, Joseph E. Gonzalez, and Ion Stoica</h5> 423 <h6><span class="conference workshop neurips">NeurIPS 2018</span> 32nd Annual Conference on Neural Information Processing Systems</h6> 424 <h6>Workshop on Systems for Machine Learning, Dec 2018</h6> 425 <p>Demonstrate 2.5x-4.9x speedups for deep learning inference workloads via GPU multitenancy</p> 426 <ul> 427 <li><a href="https://arxiv.org/abs/1901.00041">[arXiv]</a></li> 428 <li><a href="https://github.com/ucbrise/caravel/raw/master/assets/poster.pdf">[Poster]</a></li> 429 <li><a href="https://arxiv.org/abs/1901.10008">[Follow-up paper]</a></li> 430 </ul> 431 </div> 432 <img class="citation-media" src="https://d33wubrfki0l68.cloudfront.net/6e3b675ddf30988a4bbc6f3098122178b35da0e7/33b93/images/gpumulti_learningsys.jpg" alt="" width=110> 433 </section> 434 </section> 435 436 <h3>Invited talks</h3> 437 <section> 438 <ul style="font-size: 16px"> 439 <li>Oct 2022: DreamFusion: Text-to-3D using 2D Diffusion, Berkeley Computer Vision</li> 440 <li>Feb 2022: 3D content creation with data-efficient consistent neural fields, CSM.ai</li> 441 <li>Jan 2022: Data-Efficient Creative Content Creation, MIT</li> 442 <li>Oct 2021: From Prompts to Pixels: Generative Methods for AI Art, Hitachi</li> 443 <li>Sep 2021: Diffusion Probabilistic Models, BAIR Computer Vision Reading Group</li> 444 <li>Jun 2021: Putting NeRF on a Diet, ML Collective CV Paper Reading Session</li> 445 <li>2021: Putting NeRF on a Diet, MIT MIC Reading Group</li> 446 <li>Jun 2019: Oral presentation, ISCA 2019 ML For Systems</li> 447 <!-- li>NSF Graduate Research Fellowship (2019)</li--> 448 </ul> 449 </section> 450 451 <h3>Service and teaching</h3> 452 <section> 453 <ul style="font-size: 16px"> 454 <li>Seminar Coordinator, CS 294-43: Vision and Language AI Seminar, Fall 2022</li> 455 <li>Graduate Student Instructor, CS 184/284a: Computer Graphics and Imaging, Spring 2022</li> 456 <li>UC Berkeley EECS PhD Admissions Committee, 2022</li> 457 <li>Program Committee: ICML 2022, SIGGRAPH 2022, NeurIPS 2021, ICML 2021</li> 458 <li>Discussant, UAI 2020</li> 459 <li>Director, non-profit Machine Intelligence Community</li> 460 </ul> 461 </section> 462 463 <h3>
463Machine Learning Software</h3> 464 <section> 465 <ul style="font-size: 16px"> 466 <li><a href="https://github.com/google-research/google-research/tree/master/dreamfields">â Text-Guided Object Generation (JAX)</a></li> 467 <li><a href="https://github.com/parasj/contracode/">Transformers and LSTMs to represent and summarize code (PyTorch)</a></li> 468 <li><a href="https://github.com/ajayjain/DietNeRF/">â Data-efficient NeRF for 3D reconstruction and view synthesis (PyTorch)</a></li> 469 <li><a href="https://github.com/ajayjain/VectorAscent/">â Vector Ascent: Text-Guided Vector Graphics Generation (PyTorch)</a></li> 470 <li><a href="https://github.com/hojonathanho/diffusion">â Denoising Diffusion Probabilistic Models (Tensorflow)</a></li> 471 <li><a href="https://github.com/scottemmons/sgm">Long-horizon navigation with Sparse Graphical Memory (PyTorch)</a></li> 472 <li><a href="https://github.com/ajayjain/lmconv">Autoregressive image generation and outpainting (PyTorch)</a></li> 473 <li><a href="https://github.com/parasj/checkmate">â Checkmate: memory-efficient training of deep neural networks (Tensorflow)</a></li> 474 </ul> 475 </section> 476 477 <h3>More Software</h3> 478 <section> 479 <ul style="font-size: 16px"> 480 <li><a href="https://github.com/revec">LLVM revectorization</a>: <a href="https://github.com/revec/VectorBench">Benchmark</a>, <a href="https://github.com/revec/llvm-revec">â Compiler pass</a>, <a href="https://github.com/revec/intransitive">Intrinsic equivalences</a></li> 481 <li><a href="https://github.com/cauldron-desktop/cauldron-desktop">Cauldron Chrome extension for offline browsing</a> 482 </ul> 483 </section> 484 485 <footer> 486 <hr> 487 <p>ajayj at berkeley dot edu</p> 488 </footer> 489 490
490<script src="/vectorfusion/assets/js/yall.js"></script>
490 491
491<script src="https://code.jquery.com/jquery-3.5.1.slim.min.js" integrity="sha384-DfXdz2htPH0lsSSs5nCTpuj/zy4C+OGpamoFVy38MVBnE+IbbVYUew+OrCXaRkfj" crossorigin="anonymous"></script>
491 492
492<script> 493 $(document).ready(() => { 494 document.querySelectorAll('video').forEach((el, _) => { 495 el.onmouseout = el.play; 496 el.onmouseover = el.pause; 497 }); 498 499 yall( 500 { 501 observeChanges: true 502 } 503 ); 504 }); 505 </script>
505 506</body> 507</html>
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.