Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
skip to main content
Jan 2022 software

Artifact for article: PRIMA: General and Precise Neural Network Certification via Scalable Convex Hull Approximations

Description

Codebase for the ERAN verification framework including the verifier PRIMA (domains “refinepoly” or “refinegpupoly”) and scripts to reproduce the results reported in “PRIMA: General and Precise Neural Network Certification via Scalable Convex Hull Approximations”.


Assets

Read Me (popl22main-p329-p-Artifact-accepted-readme.txt)
Artifact (popl22main-p329-p-Artifact-accepted.zip)

Instructions

General Installation

Software Dependencies:

cmake (>=3.17.1) m4 (>=1.4.18) autoconf libtool pdftex

Hardware Dependencies:

To use the domain “refinegpupoly” (and reproduce our results) a recent GPU (~NVIDIA RTX 1080 Ti and above) is required.

General Installation:

Please follow the README for detailed installation instructions.

Experimental Installation

Parameterization:

Parameters are defined in the corresponding config files.

Work Flows:

After following the installation instructions navigate to the “tf_verify” directory and run “./run_experiments.sh” this will populate the “./logs” directory.

Evaluation:

Inspect the produced log files, which include a summary of their results at the bottom.


Provenance

https://github.com/eth-sri/eran


License

free


Comments