SPLASH 2022
Mon 5 - Sat 10 December 2022 Auckland, New Zealand
Wed 7 Dec 2022 13:30 - 14:00 at Lecture Theatre 2 - SAS Papers 1 Chair(s): Roberto Giacobazzi

Deep Neural Networks (DNNs) are rapidly being applied to safety-critical domains such as drone and airplane control, motivating techniques for verifying the safety of their behavior. Unfortunately, DNN verification is NP-hard, with current algorithms slowing exponentially with the number of nodes in the DNN. This paper introduces the notion of Abstract Neural Networks (ANNs), which can be used to soundly overapproximate DNNs while using fewer nodes. An ANN is like a DNN except weight matrices are replaced by values in a given abstract domain. We present a framework parameterized by the abstract domain and activation functions used in the DNN that can be used to construct a corresponding ANN. We present necessary and sufficient conditions on the DNN activation functions for the constructed ANN to soundly over-approximate the given DNN. Prior work on DNN abstraction was restricted to the interval domain and ReLU activation function. Our framework can be instantiated with other abstract domains such as octagons and polyhedra, as well as other activation functions such as Leaky ReLU, Sigmoid, and Hyperbolic Tangent.

Wed 7 Dec

Displayed time zone: Auckland, Wellington change

13:30 - 15:00
SAS Papers 1COVID Time Papers In Person at Lecture Theatre 2
Chair(s): Roberto Giacobazzi University of Verona
13:30
30m
Talk
Abstract Neural Networks
COVID Time Papers In Person
Matthew Sotoudeh Stanford University, Aditya V. Thakur University of California at Davis
Link to publication DOI
14:00
30m
Talk
Reduced Products of Abstract Domains for Fairness Certification of Neural Networks
COVID Time Papers In Person
Denis Mazzucato INRIA & École Normale Supérieure, Caterina Urban Inria & École Normale Supérieure | Université PSL
Link to publication DOI
14:30
30m
Talk
Static analysis of ReLU neural networks with tropical polyhedra
COVID Time Papers In Person
Eric Goubault Ecole Polytechnique, Sebastien Palumby Ecole Polytechnique, Sylvie Putot École Polytechnique, Louis Rustenholz Universidad Politécnica de Madrid (UPM) and IMDEA Software Institute, Sriram Sankaranarayanan University of Colorado, Boulder
Link to publication DOI