Matthew Sotoudeh
Studying Mathematics and Computer Science at the University of
California, Davis.
Currently working on a variety of problems with the
Davis Automated Reasoning Group
(DARG).
Research & Publications
-
SyReNN: A Tool for Analyzing Deep Neural Networks
Sotoudeh, M. and Thakur, A. V.
27th International Conference on Tools and Algorithms for the
Construction and Analysis of Systems (TACAS 2021, To Appear)
[Conference]
-
Analogy-Making as a Core Primitive in the Software Engineering Toolbox
Sotoudeh, M. and Thakur, A. V.
Proceedings of the 2020 ACM SIGPLAN International Symposium on
New Ideas, New Paradigms, and Reflections on Programming and
Software (SPLASH "Onward!" 2020)
[Proceedings]
[Talk]
[Poster]
[Code]
[BibTeX]
[arXiv]
-
Abstract Neural Networks
Sotoudeh, M. and Thakur, A. V.
Proceedings of the 27th Static Analysis Symposium
(SAS 2020)
[arXiv]
[Talk]
[Code]
[Conference]
-
Bounded Model Checking of Deep Neural Network Controllers
Sotoudeh, M. (Advised by Thakur, A. V.)
Won second place at the International Symposium on Principles Of
Programming Languages (POPL) 2020 Student Research Competition
[Competition]
[Code]
-
Correcting Deep Neural Networks with Small, Generalizing
Patches
Sotoudeh, M. and Thakur, A. V.
NeurIPS 2019 Workshop on Safety and Robustness in Decision Making
[PDF]
[Code]
[BibTeX]
-
Computing Linear Restrictions of Neural Networks
Sotoudeh, M. and Thakur, A. V.
Advances in Neural Information Processing Systems (NeurIPS 2019)
[Proceedings]
[Poster]
[Slides]
[Code]
[BibTeX]
[arXiv (old)]
-
ISA Mapper: A Compute and Hardware Agnostic Deep Learning Compiler
Sotoudeh, M.; Venkat, A.; Anderson, M.; Georganas, E.;
Heinecke, A.; Knight, J.
Proceedings of the 16th ACM International Conference on Computing Frontiers
(CF 2019)
[Proceedings]
[Slides]
[BibTeX]
[arXiv (old)]
-
C3-Flow: Compute Compression Co-Design Flow for Deep Neural
Networks
Sotoudeh, M. and Baghsorkhi, S. S.
Proceedings of the 56th Annual Design Automation Conference
(DAC 2019)
[Proceedings]
[Slides]
[BibTeX]
[arXiv (old)]
Conference Presentations
-
Understanding and Repairing Deep Neural Networks,
Waterloo ML + Logic Colloquium 2020, Virtual (with Aditya Thakur)
-
Abstract Neural Networks,
SAS 2020, Virtual
-
Analogy-Making as a Core Primitive in the Software Engineering Toolbox,
Onward! 2020, Virtual
-
Bounded Model Checking of Deep Neural Network Controllers,
POPL SRC 2020, New Orleans, LA
-
Computing Linear Restrictions of Neural Networks,
NeurIPS 2019, Vancouver, BC, Canada
-
ISA Mapper: A Compute and Hardware Agnostic Deep Learning Compiler,
CF 2019, Alghero, Italy
-
C3-Flow: Compute Compression Co-Design Flow for Deep Neural Networks,
DAC 2019, Las Vegas, NV
-
Correcting Deep Neural Networks with Small, Generalizing Patches,
SRDM 2019, Vancouver, BC, Canada
Honors & Awards
- Honorable Mention, CRA Outstanding Undergraduate Researcher Award
- 2nd Place, POPL 2020 Student Research Competition
- Regents Scholarship, UC Davis
- Rep. Rho Khanna Congressional Award
- Travel Awards: NeurIPS Foundation (2019), UC Davis CS Department (2020), UC Davis URC (2019)
My current email address is
masotoudeh@ucdavis.edu.
My permanent email address is
matthew@masot.net.
Neither of those accounts are entirely managed by me, so it would be
prudent to assume any communications sent to those addresses may be viewed
surreptitiously by others. You are encouraged to use my
PGP Key
if this concerns you.
Other
You can find a variety of projects on my
GitHub profile, ranging
from add-ons for Google Docs to transpilers.
I used to collect a list of
interesting proofs.
In 2017 I taught a one-week course on programming for middle schoolers
in the Bay Area. Most of the curriculum for that course, including
presentations, lesson outlines, projects, and retrospective comments
are available here.
My full name is Matthew Ali Sotoudeh, but (as far as I know) there
aren't any name collisions with my first/last alone. Hopefully it stays
that way!
I have a PDF-formatted resume.
I also have
an account
on LinkedIn. However, this website is usually significantly better
updated than either of those.