Nothing Special   »   [go: up one dir, main page]

Download PDFOpen PDF in browser

Supporting Standardization of Neural Networks Verification with VNNLIB and CoCoNet

12 pagesPublished: October 23, 2023

Abstract

The interest in the verification of neural networks has been growing steadily in recent years and there have been several advancements in theory, algorithms and tools for the verification of neural networks. Also propelled by VNNCOMP — the annual competition of tools for the verification of neural networks — the community is making steady progress to close the gap with practical applications. In this scenario, we believe that researchers and practitioners should rely on some commonly accepted standard to describe (trained) networks and their properties, as well as a toolset to visualize and to convert from common formats to such standard. The purpose of VNN-LIB and CoCoNet is precisely to provide such standard and toolset, respectively. In this paper we briefly describe the principles and design choices behind the current version of VNN-LIB standard, and we give an overview of the current and planned capabilities of CoCoNet.

Keyphrases: artificial intelligence, machine learning, neural networks verification, software engineering

In: Nina Narodytska, Guy Amir, Guy Katz and Omri Isac (editors). Proceedings of the 6th Workshop on Formal Methods for ML-Enabled Autonomous Systems, vol 16, pages 47-58.

BibTeX entry
@inproceedings{FoMLAS2023:Supporting_Standardization_Neural_Networks,
  author    = {Stefano Demarchi and Dario Guidotti and Luca Pulina and Armando Tacchella},
  title     = {Supporting Standardization of Neural Networks Verification with VNNLIB and CoCoNet},
  booktitle = {Proceedings of the 6th Workshop on Formal Methods for ML-Enabled Autonomous Systems},
  editor    = {Nina Narodytska and Guy Amir and Guy Katz and Omri Isac},
  series    = {Kalpa Publications in Computing},
  volume    = {16},
  publisher = {EasyChair},
  bibsource = {EasyChair, https://easychair.org},
  issn      = {2515-1762},
  url       = {/publications/paper/Qgdn},
  doi       = {10.29007/5pdh},
  pages     = {47-58},
  year      = {2023}}
Download PDFOpen PDF in browser