This program is tentative and subject to change.

Sun 20 Oct 2024 14:00 - 15:00 at San Gabriel - Machine learning and Neural networks Chair(s): Marco Campion

The tutorial will guide participants through the use of abstract interpretation-based static analysis for the certification of hyperproperties of machine learning models. A neural network surrogate from a real-world avionics use case will serve as a running example throughout the session. The design of the static analysis will proceed systematically, beginning from the formal semantics definitions tailored to the property of interest, an instance of abstract non-interference. We will then overview various existing (combinations of) abstractions, exemplifying some of their lesser-known characteristics. Finally, we will address practical implementation challenges, ensuring a comprehensive demonstration that abstract interpretation can effectively be used to certify critical properties of high-stakes machine learning applications.

This program is tentative and subject to change.

Sun 20 Oct

Displayed time zone: Pacific Time (US & Canada) change

14:00 - 15:30
Machine learning and Neural networksSAS at San Gabriel
Chair(s): Marco Campion INRIA & École Normale Supérieure | Université PSL
14:00
60m
Tutorial
Abstract Interpretation-Based Certification of Hyperproperties for High-Stakes Machine Learning Software
SAS
Caterina Urban Inria & École Normale Supérieure | Université PSL
15:00
30m
Full-paper
Robustness Verification of Multi-Label Neural Network Classifiers
SAS