PFIA 2020

Every year or so, I have an unexplicable urge to do a tutorial. Be it for undergraduate students or seasoned researchers, I enjoy teaching and learning from the experience. So this time, I wanted to show how we could verify a neural network, with real tools. I designed a toy problem and trained a neural network to solve it; the goal of this tutorial was to formally verify that the model was respecting safety properties.

As always, materials are available on the top widget (code and slides are on an open INRIA gitlab repository). A replay of the tutorial is available here (it was scorching hot in my room at the moment, so I had to stop sometimes to cath my breath).

Researcher on Trustworthy Artificial Intelligence