-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathindex.html
executable file
·107 lines (107 loc) · 6.11 KB
/
index.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
<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.1//EN"
"http://www.w3.org/TR/xhtml11/DTD/xhtml11.dtd">
<html xmlns="http://www.w3.org/1999/xhtml" xml:lang="en">
<head>
<link rel="icon" type="image/png" href="https://www.charleslo.net/favicon.png">
<meta name="generator" content="jemdoc, see http://jemdoc.jaboc.net/" />
<meta http-equiv="Content-Type" content="text/html;charset=utf-8" />
<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/font-awesome/4.7.0/css/font-awesome.min.css">
<link rel="stylesheet" href="https://cdn.rawgit.com/jpswalsh/academicons/master/css/academicons.min.css">
<!-- Global site tag (gtag.js) - Google Analytics -->
<script async src="https://www.googletagmanager.com/gtag/js?id=UA-55937454-2"></script>
<script>
window.dataLayer = window.dataLayer || [];
function gtag(){dataLayer.push(arguments);}
gtag('js', new Date());
gtag('config', 'UA-55937454-2');
</script>
<link rel="stylesheet" href="jemdoc.css" type="text/css" />
<title>Charles Lo</title>
</head>
<body>
<table summary="Table for page layout." id="tlayout">
<tr valign="top">
<td id="layout-menu">
<div class="menu-category">MAIN</div>
<div class="menu-item"><a href="index.html" class="current">home</a></div>
<div class="menu-item"><a href="pubs.html">publications</a></div>
<div class="menu-item"><a href="pubs/charleslo_cv.pdf">cv</a></div>
</td>
<td id="layout-content">
<div id="toptitle">
<h1>Charles Lo | <a href="https://www.linkedin.com/in/charles-lo/" class="fa fa-linkedin"></a> <a href="https://www.github.com/charleslo/" class="fa fa-github"></a> <a href="https://scholar.google.ca/citations?user=sxzUvqoAAAAJ" class="ai ai-google-scholar"></a> </h1>
</div>
<p>Digital Hardware and Software Engineer<br />
e-mail: charles at charleslo dot net<br /></p>
<h1>Research</h1>
<p>I am interested in hardware design, reconfigurable computing and machine learning. In the past I have designed high-performance and multi-FPGA accelerators for Object Detectors and neural networks. I was also involved with the early development of the Xilinx SDAccel tools for integrating FPGA accelerators in the OpenCL heterogeneous computing framework. Most recently, I have been applying Gaussian processes in Bayesian optimization techniques to perform design-space exploration of High-Level Synthesis and other hardware generators.</p>
<h1>Other Projects</h1>
<h2>Flexible Gaussian process Library <a href="https://github.com/UofT-HPRC/hiergp" target=“blank”>GitHub</a></h2>
<ul>
<li><p>Python 3 software used in research to describe hierarchical Gaussian process models </p>
</li>
<li><p>Supports composing squared exponential and linear kernels</p>
</li>
<li><p>Uses Cython-based fast exponential library to compute squared exponential kernel</p>
</li>
</ul>
<h2>“Image Labelling using Feature Learning and Boltzmann Machine-Augmented CRFs,” ECE1510 Project Report, 2014<br /> <a href="https://www.charleslo.net/pubs/lo.ece1510.pdf" target=“blank”>PDF</a></h2>
<table class="imgtable"><tr><td>
<img src="ece1510.png" alt="" width="550px" /> </td>
<td align="left"></td></tr></table>
<ul>
<li><p>Evaluated the combination of Neural Networks, Conditional Random Fields (CRFs) and Restricted Boltzmann Machines for image labelling.</p>
</li>
<li><p>Image segmentation was performed to obtain superpixels which were inferred using a neural network.</p>
</li>
<li><p>A CRF was then used to smooth labelling across adjacent superpixels while a global RBM provided location-based labelling.</p>
</li>
<li><p>Results showed the benefit of combining the techniques, but superpixel-based classification held back performance.</p>
</li>
</ul>
<h2>“Heterogeneous Stream Computing in SAVI,” ECE1548 Project Report, 2013<br /> <a href="http://www.charleslo.net/pubs/lo.ece1548.pdf" target=“blank”>PDF</a></h2>
<table class="imgtable"><tr><td>
<img src="ece1548.png" alt="" /> </td>
<td align="left"></td></tr></table>
<ul>
<li><p>Proposed a method of mapping streaming task graphs on to virtualized heterogeneous resources in a cloud environment.</p>
</li>
<li><p>Compute kernel management and routing inspired by Software Defined Networking to simplify global control.</p>
</li>
<li><p>Preliminary prototype designed with x86 virtual machines, virtualized FPGA kernels and OpenFlow. </p>
</li>
</ul>
<h2>“A High-Performance Architecture for Training Viola-Jones Object Detectors,” MASc Project, 2012 <a href="https://tspace.library.utoronto.ca/bitstream/1807/33294/3/Lo_Charles_K_201211_MASc_thesis.pdf" target=“blank”>Thesis</a>, <a href="https://ieeexplore.ieee.org/abstract/document/6412131" target=“blank”>Paper</a></h2>
<table class="imgtable"><tr><td>
<img src="vjaccel.png" alt="" width="400px" /> </td>
<td align="left"></td></tr></table>
<ul>
<li><p>Training an ensemble of decision trees is highly task parallel but not well suited for GPUs</p>
</li>
<li><p>Developed a PCIe-FPGA system targeting a Xilinx Virtex-6 device to accelerator this task using a systolic array architecture</p>
</li>
<li><p>Performance of the floorplanned array scaled linearly and outperformed the multi-threaded OpenCV implementation</p>
</li>
</ul>
<h2>“Nonlinear Dimensionality Reduction for Music Feature Extraction,” CSC2515 Project Report, 2010<br /> <a href="http://www.charleslo.net/pubs/lo.csc2515.pdf" target=“blank”>PDF</a></h2>
<table class="imgtable"><tr><td>
<img src="csc2515.png" alt="2-D Visualization of Compressed Features" width="549px" height="327px" /> </td>
<td align="left"></td></tr></table>
<ul>
<li><p>Experimented with PCA, Autoencoders, LLE and t-SNE as methods for compressing high-dimensional audio features for music genre classification.</p>
</li>
<li><p>Compression would allow for short feature tags to group together similar types of music in large databases.</p>
</li>
<li><p>Results found the t-SNE performed the best in maintaining neighbourhood structure.</p>
</li>
</ul>
<div id="footer">
<div id="footer-text">
Page generated 2020-07-25 23:38:28 CEST, by <a href="https://github.com/wsshin/jemdoc_mathjax" target="blank">jemdoc+MathJax</a>.
</div>
</div>
</td>
</tr>
</table>
</body>
</html>