-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathml.html
214 lines (207 loc) · 11.8 KB
/
ml.html
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
---
title: Machine Learning
subtitle: Lift as a cross-platform ML engine
layout: project_page
permalink: /ml
---
<div id="main">
<div class="container">
<div class="row">
<!-- Content -->
<div id="content" class="8u skel-cell-important">
<section>
<header>
<h2>Posters</h2>
</header>
<div class="row">
<section class="6u">
<div class="thumb_frame">
<a href="posters/2019/MogersHiPEAC2019Poster.pdf">
<div class="thumb_container">
<img class="thumb_image" src="posters/2019/thumbnails/MogersHiPEAC2019Poster_thumb.png"
width="300" height="212">
<div class="thumb_overlay">
<div class="thumb_text">
<p class="posted">January 21, 2019</p>
<br>
<p class="posted">@ the EDLA 2019 workshop at HiPEAC</p>
<br>
<p>Towards Mapping Lift to Deep Neural Network Accelerators</p>
</div>
</div>
</div>
</a>
</div>
</section>
<section class="6u">
<div class="thumb_frame">
<a href="posters/2017/MogersPPar2017Poster.png">
<div class="thumb_container">
<img class="thumb_image" src="posters/2017/thumbnails/MogersPPar2017Poster_thumb.png"
width="300" height="212">
<div class="thumb_overlay">
<div class="thumb_text">
<p class="posted">June 14, 2017</p>
<br>
<p class="posted">@ the PPar Student Showcase Event</p>
<br>
<p>Optimization of neural computations using a functional data-parallel language</p>
</div>
</div>
</div>
</a>
</div>
</section>
</div>
</section>
<section>
<header>
<h2>Abstract</h2>
</header>
<p>Performance-portable code is hard to produce due to diversity and
heterogeneity of the state-of-the-art hardware platforms. Even more
complex is the task of optimizing Artificial Neural Networks (ANNs)
towards multiple hardware platforms. Manual optimization is expensive,
while modern automated tools either support a narrow set of platforms or
do not exploit individual strengths of different platforms to the fullest.</p>
<p>The functional data-parallel language <span class="lift-bold">Lift</span> was shown to be performance-portable;
the performance of the compiled OpenCL code is on par or better than that of highly
tuned platform-specific libraries. This project aims to extend the method to the
domain of Artificial Neural Networks by integrating domain-specific optimisations
into the rewrite rules-based <span class="lift-bold">Lift</span> compiler.</p>
</section>
<section>
<div class="row">
<header>
<h2>Optimizational methods of interest</h2>
</header>
<div class="6u">
<ul>
<li>Parallel mappings space exploration</li>
<li>Memory tiling</li>
<li>Memory coalescing</li>
<li>Approximate computations</li>
<li>Float quantization</li>
<li>Neuron pruning</li>
<li>Training batch size autotuning</li>
</ul>
</div>
<div class="6u">
<ul>
<li>Varying precision across layers and neurons</li>
<li>Convolution kernel decomposition</li>
<li>Sharing 32-bit registers</li>
<li>OpenCL kernel fusion</li>
<li>Expression simplification</li>
<li>Proprietary instruction sets usage</li>
</ul>
</div>
</div>
<div class="row">
<header>
<h2>Publications</h2>
</header>
<ul>
<li>
Naums Mogers, Lu Li, Valentin Radu, and Christophe Dubach:
<strong><a href="https://naumsmogers.me/assets/other/CC22_liftpar.pdf">
Mapping Parallelism in a Functional IR through Constraint Satisfaction: A Case Study on Convolution for Mobile GPUs
</a></strong>;
<a href="https://conf.researchr.org/home/CC-2022">
ACM SIGPLAN 2022 International Conference on Compiler Construction
</a>
</li>
<li>
Naums Mogers, Valentin Radu, Lu Li, Jack Turner, Michael O'Boyle, and Christophe Dubach:
<strong><a href="publications/2020/mogers2020directconvonmobile.pdf">
Automatic Generation of Specialized Direct Convolutions for Mobile GPUs
</a></strong>;
<a href="https://insight-archlab.github.io/gpgpu.html">
General Purpose Processing Using GPU (GPGPU) 2020
</a>
</li>
<li>
Naums Mogers, Aaron Smith, Dimitrios Vytiniotis, Michel Steuwer, Christophe Dubach, Ryota Tomioka:
<strong><a href="publications/2019/mogers2019dnnaccelerators.pdf">
Towards Mapping Lift to Deep Neural Network Accelerators
</a></strong>;
<a href="https://www.hipeac.net/2019/valencia/#/schedule/sessions/7627/">
Workshop on Emerging Deep Learning Accelerators (EDLA) 2019 @ HiPEAC
</a>
</li>
</ul>
</div>
</section>
</div>
<!-- Sidebar -->
<div id="sidebar" class="4u">
<section>
<header>
<h2>Talks</h2>
</header>
<div class="row">
<section>
<ul class="style">
<li>
<p class="posted">January 21st, 2019 @ the EDLA 2019 workshop at HiPEAC</p>
<p><a href="presentations/2019/EDLA-2019.pdf">
Towards Mapping Lift to Deep Neural Network Accelerators</a></p>
</li>
<li>
<p class="posted">June 14th, 2017 @ the PPar Student Showcase Event</p>
<p><a href="presentations/2017/MogersPPar2017.pdf">
Optimization of neural computations using a functional data-parallel language</a></p>
</li>
</ul>
</section>
</div>
</section>
<section class="profile">
<header>
<h2>Researchers</h2>
</header>
<div class="row">
<section class="6u">
<a href="https://naumsmogers.me/" class="image full">
<img src="images/naums.jpg" alt="Naums Mogers"></a>
<a href="https://naumsmogers.me/">Naums Mogers</a>
<br>
PhD Student
<br>
<a href="http://www.ed.ac.uk/informatics/">University of Edinburgh</a>
</section>
<section class="6u">
<a href="http://www.inf.ed.ac.uk/people/staff/Lu_Li.html" class="image full">
<img src="images/lu.jpg" alt="Lu Li"></a>
<a href="http://www.inf.ed.ac.uk/people/staff/Lu_Li.html">Lu Li</a>
<br>
Postdoctoral Researcher
<br>
<a href="http://www.ed.ac.uk/informatics/">University of Edinburgh</a>
</section>
</div>
<div class="row">
<section class="6u">
<a href="https://michel-steuwer.github.io/" class="image full">
<img src="images/msteuwer.jpg" alt="Michel Steuwer"></a>
<a href="https://michel-steuwer.github.io/">Michel Steuwer</a>
<br>
Lecturer
<br>
<a href="https://www.gla.ac.uk/schools/computing/">University of Glasgow</a>
</section>
<section class="6u">
<a href="http://homepages.inf.ed.ac.uk/cdubach/" class="image full">
<img src="images/cdubach.png" alt="Christophe Dubach"></a>
<a href="http://homepages.inf.ed.ac.uk/cdubach/">Christophe Dubach</a>
<br>
Reader
<br>
<a href="http://www.ed.ac.uk/informatics/">University of Edinburgh</a>
</section>
</div>
</section>
</div>
</div>
</div>
</div>