NEUROSPF: A Tool for the Symbolic Analysis of Neural Networks

Muhammad Usman, Yannic Noller, Corina S. Pasareanu, Youcheng Sun, Divya Gopinath

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

This paper presents NEUROSPF, a tool for the symbolic analysis of neural networks. Given a trained neural network model, the tool extracts the architecture and model parameters and translates them into a Java representation that is amenable for analysis using the Symbolic PathFinder symbolic execution tool. Notably, NEUROSPF encodes specialized peer classes for parsing the model's parameters, thereby enabling efficient analysis. With NEUROSPF the user has the flexibility to specify either the inputs or the network internal parameters as symbolic, promoting the application of program analysis and testing approaches from software engineering to the field of machine learning. For instance, NEUROSPF can be used for coverage-based testing and test generation, finding adversarial examples and also constraint-based repair of neural networks, thus improving the reliability of neural networks and of the applications that use them. Video URL: https://youtu.be/seal8fG78L.

Original languageEnglish
Title of host publicationProceedings - 2021 IEEE/ACM 43rd International Conference on Software Engineering
Subtitle of host publicationCompanion Proceedings, ICSE-Companion 2021
PublisherIEEE Computer Society
Pages25-28
Number of pages4
ISBN (Electronic)9781665412193
DOIs
Publication statusPublished - May 2021
Event43rd IEEE/ACM International Conference on Software Engineering: Companion, ICSE-Companion 2021 - Virtual, Online, Spain
Duration: 25 May 202128 May 2021

Publication series

NameProceedings - International Conference on Software Engineering
ISSN (Print)0270-5257

Conference

Conference43rd IEEE/ACM International Conference on Software Engineering: Companion, ICSE-Companion 2021
Country/TerritorySpain
CityVirtual, Online
Period25/05/2128/05/21

Keywords

  • Neural Networks
  • Symbolic Execution
  • Symbolic PathFinder

Fingerprint

Dive into the research topics of 'NEUROSPF: A Tool for the Symbolic Analysis of Neural Networks'. Together they form a unique fingerprint.

Cite this