PhD Seminar • Artificial Intelligence • Reinforcement Learning Fuzz Testing for Hard Combinatorial ProblemsExport this event to calendar

Tuesday, November 1, 2022 — 4:00 PM to 5:00 PM EDT

Please note: This PhD seminar will take place online.

Joseph Scott, PhD candidate
David R. Cheriton School of Computer Science

Supervisor: Professor Vijay Ganesh

Solvers for hard combinatorial problems, such as SAT/SMT Solvers, are infamous for underlying worst-case complexity, despite industrial success. For example, the cvc5 SMT solver can solve problems of over a gigabyte in size from certain benchmark families while failing to solve benchmarks that can fit on a single screen. In the latter case, when a competing solver (e.g., z3) solves said benchmark quickly, it suggests the slowdown is caused by programming/solver error rather than the computational complexity of the problem at hand.

To this end, we present two tools. First, BanditFuzz is a performance fuzzer for SMT Solvers. BanditFuzz implements a reinforcement learning loop to find benchmarks where a target solver is slow while a set of relative solvers is fast. BanditFuzz leverages the natural feedback collected from running the solvers to compute a strong reward signal to achieve its goal. Finally, we present Pierce, an upcoming fuzzing tool for neural network verification.


To join this PhD seminar on Zoom, please go to https://uwaterloo.zoom.us/j/98165589716.

Location 
Online PhD seminar
200 University Avenue West

Waterloo, ON N2L 3G1
Canada
Event tags 

S M T W T F S
25
26
27
28
29
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
1
2
3
4
5
6
  1. 2024 (100)
    1. April (23)
    2. March (27)
    3. February (25)
    4. January (25)
  2. 2023 (296)
    1. December (20)
    2. November (28)
    3. October (15)
    4. September (25)
    5. August (30)
    6. July (30)
    7. June (22)
    8. May (23)
    9. April (32)
    10. March (31)
    11. February (18)
    12. January (22)
  3. 2022 (245)
  4. 2021 (210)
  5. 2020 (217)
  6. 2019 (255)
  7. 2018 (217)
  8. 2017 (36)
  9. 2016 (21)
  10. 2015 (36)
  11. 2014 (33)
  12. 2013 (23)
  13. 2012 (4)
  14. 2011 (1)
  15. 2010 (1)
  16. 2009 (1)
  17. 2008 (1)