Obsidian/Notes on Papers/A Review of Formal Methods applied to Machine Learning.md

1.0 KiB

First Pass

Category: This is a review paper.

Context: This paper tries to keep up with where formal methods are for machine learning based systems. It is not necessarily a controls paper, but a broader machine learning paper in general.

Correctness: Really well written on the first pass. Easy to understand and things seem well cited.

Contributions: Great citations showing links to different papers and also provides nice spots for forward research. Talks about how verification needs to be done along the whole pipeline: from data prep to training to implementation. There needs to be more work on proving things about model behavior, but in general this review has a positive outlook on the field.

Clarity: Very well written, easy to understand. Except, what is abstractification of a network?

Second Pass

What is the main thrust?

What is the supporting evidence?

What are the key findings? Local robustness

Third Pass

Recreation Notes:

Hidden Findings:

Weak Points? Strong Points?