https://armingerami.github.io/

1<!DOCTYPE html>
2<html lang="en">
3  <head>
4    <meta charset="UTF-8" />
5    <link rel="icon" type="image/svg+xml" href="favicon.svg" />
6    <link rel="stylesheet" href="style.css" />
7    <meta name="viewport" content="width=device-width, initial-scale=1.0" />
8    <title>Armin's Website</title>
9  </head>
10  <body>
11
12    <canvas id="bg"></canvas>   
13
14    <main>
15
16      <header>
17        <h1>Armin Gerami</h1>
18        <h6>[email protected]</h6>
19        <p>🔥 Welcome to my webpage!</p>
20      </header>
21
22
23      <!-- <blockquote>
24        <p>I'm a Computer Science PhD student at University of Maryland, College Park, and advised by prof. Ramani Duraiswami. My research interests are scientific ML and differetiable programming. </p>
25      </blockquote> -->
26
27      <section>
28        <h2>📜 Bio</h2>
29        <p>
30          I'm a Computer Science Ph.D. candidate at the University of Maryland, working under the guidance of <a href="https://www.cs.umd.edu/people/rduraisw" style="color: rgb(211,211,211);"><u>Prof. Ramani Duraiswami</u></a>. My research interests lie in the domains of high performane computing, deep learning, and differentiable programming. Currently, I'm working on improving the computational efficiency of Transformers, Linear layres in neural networks, and Spatial Audio Rendering. 
31        </p>
32      </section>
33
34      <section class="light">
35        <h2>📚 Education</h2>
36        <p>
37          Ph.D., Computer Science, University of Maryland, 2023 - Present.
38        </p>
39        <p>
40          M.Sc., Electrical Engineering, University of Maryland, 2022 - 2023.
41        </p>
42        <p>
43          B.Sc., Electrical Engineering, Sharif University of Technology, 2016 - 2020.
44        </p>
45
46        <h2>🏆 Awards</h2>
47        <p>
48          Recipient of the NSF NeuroPAC Fellowship Award for Summer 2025.
49        </p>
50        <p>
51          Outstanding graduate research assistant awrad for AY 2023-2024.
52        </p>
53        <p>
54          Ranked 21st in Iran's National University Entrance Exam (Konkour) in 2016.
55        </p>
56
57      </section>
58
59      <section class="left">
60        <h2>🔬 Research</h2>
61        <p>
62          <i>Auditing Algorithmic Bias in Transformer-Based Stock Trading</i><br>
63          <span style="color: rgb(121, 55, 179);"><b>Armin Gerami</b></span>, Ramani Duraiswami; <span style="color: rgb(217, 53, 26);"><b>Neurips'25</b></span><br>
64          <b><span style="color: rgb(90, 117, 161);">Summary</span>:</b> This study introduces a metric based on Partial Information Decomposition (PID) to audit how transformer models used in finance rely on volatile data and are affected by price movement frequency. The analysis reveals that the model entirely disregards data volatility and shows a significant bias toward assets with lower-frequency price movements.
65        </p>
66        <p>
67          <i>Optimized Linear Attention GPU Kernel Implementation</i><br>
68          <span style="color: rgb(121, 55, 179);"><b>Armin Gerami</b></span>, Ramani Duraiswami; <span style="color: rgb(217, 53, 26);"><b>TMLR'25</b></span><br>
69          <b><span style="color: rgb(90, 117, 161);">Summary</span>:</b> Linear attention (Kernel Separation) is a promising approach for calculating the attention in transformers with linear time complexity. However, it introduces a significant data movement overhead. We present a CUDA library that achieves a 3.3× speedup and 3.6× reduction in memory consumption for linear attention implementation compared to the state-of-the-art.
70        </p>
71        <p>
72          <i>GUST: Graph Edge-Coloring Utilization for Accelerating Sparse Matrix Vector Multiplication</i><br>
73          <span style="color: rgb(121, 55, 179);"><b>Armin Gerami</b></span>, Bahar Asgari; <span style="color: rgb(217, 53, 26);"><b>ASPLOS'25</b></span><br>
74          <b><span style="color: rgb(90, 117, 161);">Summary</span>:</b> A software/hardware codesign to accelerate sparse matrix-vector multiplication. The harware enables resource sharing while the graph-edge-coloring schedules the input stream to prevent collisions.
75        </p>
76        <p>
77          <i>Efficient Spatial Audio Rendering Via Differentiable FIR To IIR Estimation</i><br>

78          <span style="color: rgb(121, 55, 179);"><b>Armin Gerami</b></span>, Bowen Zhi, Dmitry N. Zotkin, Ramani Duraiswami; <span style="color: rgb(217, 53, 26);"><b>ICASSP'25</b></span><br>
79          <b><span style="color: rgb(90, 117, 161);">Summary</span>:</b> We introduce a novel convex optimization algorithm for estimating FIR filters with an IIR filter, and achieve 2× speedup with an accuracy of %0.01 through differential programming.
80        </p>
81        <p>
82          <i>Towards Efficient Implementation of Differentiable Sparse Matrix Multiplication</i><br>
83          <span style="color: rgb(121, 55, 179);"><b>Armin Gerami</b></span>, Ramani Duraiswami; <b>Preprint</b><br>
84          <b><span style="color: rgb(90, 117, 161);">Summary</span>:</b>  Sparse Matrix-Matrix Multiplication (SpMM) is a crucial operation for improving the efficiency of large deep neural networks, but creating a high-performance, differentiable version compatible with gradient-based training is challenging. This paper introduces a novel differentiable SpMM implementation that achieves a 2.6x speedup over the state-of-the-art, enabling more efficient and scalable training for sparse neural network models.
85        </p>
86        <p>
87          <i>Room Impulse Response Synthesis via Differentiable Feedback Delay Networks</i><br>
88          <span style="color: rgb(121, 55, 179);"><b>Armin Gerami</b></span>, Ramani Duraiswami; <b>Preprint</b><br>
89          <b><span style="color: rgb(90, 117, 161);">Summary</span>:</b>  We introduce a computationally efficient and tunable feedback-delay-network (FDN) architecture for real-time room-impulse-response (RIR) rendering, addressing the computational and latency challenges inherent in traditional convolution and Fourier transform-based methods.
90        </p>
91        
92
93        <h2>🚀 Invention Disclosures</h2>
94        <p>
95          Differentiable FIR To IIR Filter Estimation
96        </p>
97        <p>
98          Rapid Energy and Emission Auditor
99        </p>
100                
101
102        <h2>💻 Coding Languages</h2>
103        <p>
104          CUDA, C++, Python, Verilog
105        </p>
106      </section>
107
108      <section class="light">
109        <h2>(Bonus) Puzzle:</h2>
110        <p>
111          You have a rope that is 150 feet long. You can tie it in different ways and can also cut it. You are on a cliff that is 200 feet tall. 100 feet from the top is a ledge. There are rings to tie the rope to at the top and at the ledge. How can you get down?<br>
112        </p>
113      </section>
114
115<section class="light">
116        <h2>(Bonus Bonus) Chess Puzzle:</h2>
117<iframe id="12257825" allowtransparency="true" frameborder="0" style="width:100%;border:none;" src="https://www.chess.com/emboard?id=12257825"></iframe>

117<script>window.addEventListener("message",e=>{e['data']&&"12257825"===e['data']['id']&&document.getElementById(`${e['data']['id']}`)&&(document.getElementById(`${e['data']['id']}`).style.height=`${e['data']['frameHeight']+37}px`)});</script>

117
118      </div>
119
120    <!-- 

120<script type="text/javascript">
121    var request = new XMLHttpRequest();
122    request.open('GET', 'https://chesspuzzle.net/Daily/Api', true);
123    request.onload = function () {
124    if (request.status >= 200 && request.status < 400) {
125    var result = JSON.parse(request.responseText);
126    document.getElementById("puzzleText").textContent = result.Text;
127    document.getElementById("puzzleLink").href = result.Link;
128    document.getElementById("puzzleImage").src = result.Image;
129    }
130    };
131    request.send();
132    </script>

132 -->
133    <!-- <iframe style="width: 400px; height: 560px;" src="https://www.chess.com/emboard?id=12257825" frameborder="0"></iframe>
134      </section> -->
135
136<!--       <blockquote>
137        <a href="https://www.linkedin.com/in/armingerami/"><img src="./favicon.svg" alt="Logo"></a>
138      </blockquote> -->
139
140<section_trans>
141<p><a href="https://www.linkedin.com/in/armingerami/"><img src="./linkedin.png" alt="Logo" width="80" height="80"></a>&nbsp&nbsp&nbsp&nbsp&nbsp&nbsp
142 <a href="https://github.com/armingerami"><img src="./github.png" alt="Logo" width="80" height="80"></a>&nbsp&nbsp&nbsp&nbsp&nbsp&nbsp
143 <a href="https://scholar.google.com/citations?view_op=list_works&hl=en&user=XKrFIXMAAAAJ"><img src="./scholar.png" alt="Logo" width="80" height="80"></a>&nbsp&nbsp&nbsp&nbsp&nbsp&nbsp
144 <a href="[email protected]"><img src="./email.png" alt="Logo" width="100" height="80"></a></p>
145</section_trans>
146    </main>
147    
148    

148<script type="module" src="./main.js"></script>

148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166  </body>
167</html>
168
169
170
171
172
173
174
175
176
177
178
179