<html xmlns:o="urn:schemas-microsoft-com:office:office" xmlns:w="urn:schemas-microsoft-com:office:word" xmlns:m="http://schemas.microsoft.com/office/2004/12/omml" xmlns="http://www.w3.org/TR/REC-html40">
<head>
<meta http-equiv="Content-Type" content="text/html; charset=us-ascii">
<meta name="Generator" content="Microsoft Word 15 (filtered medium)">
<style><!--
/* Font Definitions */
@font-face
{font-family:"Cambria Math";
panose-1:2 4 5 3 5 4 6 3 2 4;}
@font-face
{font-family:Calibri;
panose-1:2 15 5 2 2 2 4 3 2 4;}
/* Style Definitions */
p.MsoNormal, li.MsoNormal, div.MsoNormal
{margin:0in;
font-size:11.0pt;
font-family:"Calibri",sans-serif;
mso-ligatures:standardcontextual;}
span.EmailStyle17
{mso-style-type:personal-compose;
font-family:"Calibri",sans-serif;
color:windowtext;}
.MsoChpDefault
{mso-style-type:export-only;
font-family:"Calibri",sans-serif;}
@page WordSection1
{size:8.5in 11.0in;
margin:1.0in 1.0in 1.0in 1.0in;}
div.WordSection1
{page:WordSection1;}
--></style>
</head>
<body lang="EN-US" link="#0563C1" vlink="#954F72" style="word-wrap:break-word">
<div class="WordSection1">
<p class="MsoNormal">This is an announcement of Maksim Levental's Candidacy Exam.<o:p></o:p></p>
<p class="MsoNormal">===============================================<o:p></o:p></p>
<p class="MsoNormal">Candidate: Maksim Levental<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">Date: October 23,2023<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">Time: 2:00pm-3:00pm<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">Remote Location: <o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">Location: JCL 298<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">Title: TOWARDS AN ABSTRACTION-FREE PROGRAMMING MODEL FOR DEEP<o:p></o:p></p>
<p class="MsoNormal">LEARNING ACCELERATORS<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">Abstract: The thesis of this dissertation is that it is possible to design a programming model for expert ML developers to both represent Deep Neural Networks simultaneously at multiple levels of abstraction and generate optimal, target-specific,
code for novel accelerator architectures. Specifically, that by leveraging and extending an existing compiler, we can provide both an embedded domain specific language, including tensor, vector, scalar operations, concurrency, memory hierarchy, explicit data
movement, and we can perform optimal code generation for novel architectures. This thesis is derived of experience over the course of several projects (see Chapter 1), each of which involved manipulating highly abstract APIs and struggling to achieve runtime
performance. We validate this thesis by leveraging and extending MLIR to provide both an embedded domain specific language and optimal code generation for AMD XDNA spatial architectures. We evaluate our framework by designing, optimizing, and benchmarking
several Large Language Models.<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">Advisors: Ian Foster<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">Committee Members: Ian Foster, Kyle Chard ,Raul Fernandez, & Stephen Neuendorffer<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
</div>
</body>
</html>