Title:
Improving abstract propagation for verification of neural networks
Authors:
- Stefano Demarchi
- Andrea Gimelli
- Armando Tacchella
Published in:
(2024). ECMS 2024, 38th Proceedings
Edited by: Daniel Grzonka, Natalia Rylko, Grazyna Suchacka, Vladimir Mityushev, European Council for Modelling and Simulation.
DOI: http://doi.org/10.7148/2024
ISSN: 2522-2422 (ONLINE)
ISSN: 2522-2414 (PRINT)
ISSN: 2522-2430 (CD-ROM)
ISBN: 978-3-937436-84-5
ISBN: 978-3-937436-83-8 (CD) Communications of the ECMS Volume 38, Issue 1, June 2024, Cracow, Poland June 4th – June 7th, 2024
DOI:
https://doi.org/10.7148/2024-0483
Citation format:
Stefano demarchi, Andrea gimelli, Armando tacchella (2024). Improving Abstract Propagation for Verification of Neural Networks, ECMS 2024, Proceedings Edited by: Daniel Grzonka, Natalia Rylko, Grazyna Suchacka, Vladimir Mityushev, European Council for Modelling and Simulation. doi:10.7148/2024-0483
Abstract:
Formal verification of neural networks is a crucial technique to increase their dependability in safety critical applications. In this paper we address some scalability challenges in our verification tool NEVER2 by proposing strategies to enhance speed and overall performances. First, we apply a precomputation technique based on symbolic bounds propagation in order to improve the network analysis by determining neuron stability a priori. Second, we combine the strengths of different levels of abstraction towards a refinement strategy. We experiment with the proposed techniques on some verification benchmarks from the annual competition of verification tools for neural networks (VNN-COMP).
Full text:
Download full text