Benutzer: Gast  Login
Titel:

Automatic Abstraction Refinement in Neural Network Verification using Sensitivity Analysis

Dokumenttyp:
Konferenzbeitrag
Art des Konferenzbeitrags:
Vortrag / Präsentation
Autor(en):
Ladner, Tobias; Althoff, Matthias
Abstract:
The formal verification of neural networks is essential for their application in safety-critical environments. However, the set-based verification of neural networks using linear approximations often obtains overly conservative results, while nonlinear approximations quickly become computationally infeasible in deep neural networks. We address this issue for the first time by automatically balancing between precision and computation time without splitting the propagated set. Our work intro...     »
Stichworte:
Neural networks, formal verification, automatic abstraction refinement, sensitivity analysis, set-based computing, polynomial zonotopes
Dewey-Dezimalklassifikation:
000 Informatik, Wissen, Systeme
Kongress- / Buchtitel:
Proceedings of the 26th ACM International Conference on Hybrid Systems: Computation and Control (HSCC)
Datum der Konferenz:
09.05.2023
Verlag / Institution:
ACM
Publikationsdatum:
09.05.2023
Jahr:
2023
Reviewed:
ja
Sprache:
en
Erscheinungsform:
WWW
Volltext / DOI:
doi:10.1145/3575870.3587129
WWW:
https://dl.acm.org/doi/abs/10.1145/3575870.3587129
TUM Einrichtung:
School of Computation, Information and Technology
 BibTeX