Verified Deep Learning [CS 839 / S20]
Aws Albarghouthi
Office hrs: Tue, 1300-1415, CS 6363
TA: Swati Anand, sanand24@wisc.edu

Deep neural networks are fragile and their behaviors are often surprising. In many settings, we need to provide formal guarantees on the safety, security, correctness, or robustness of neural networks. This course covers foundational ideas from formal verification and their application to deep learning.



The couse is based on chapters from the instructor's ongoing book on the subject and research papers from machine learning and verification.

1.20 Overview of verified deep learning
Tue  chapter 1 / slides
Thu  chapter 2

1.27 Correctness properties of neural networks
Tue  chapter 3 / Hoare's classic paper / Floyd's paper / Turing's obscure paper
Thu  chapter 4

2.3 Logical encodings of neural networks
Tue 
Thu 

2.10 TBD
Tue 
Thu 

2.17 TBD
Tue  ✈️ Aws out of town for a PC meeting on Tuesday
Thu 

3.2 TBD
Tue 
Thu 

3.9 TBD
Tue 
Thu 

3.16 🌷Spring break

3.23 TBD
Tue 
Thu 

3.30 TBD
Tue 
Thu 

4.6 TBD
Tue 
Thu 

4.13 TBD
Tue 
Thu 

4.20 TBD
Tue 
Thu 

4.27 TBD
Tue 
Thu 



Grade distribution
75% project
25% assignments

Project details
Find project partner and choose project
Create GitHub repository in CS839 organization
Weekly update README.md with new developments, problems, etc.