There is a newer version of the record available.

Published January 10, 2022 | Version 1
Software Open

LinSyn: Synthesizing Tight Linear Bounds for Arbitrary Neural Network Activation Functions

  • 1. University of Southern California

Description

This artifact contains the code for the paper LinSyn: Synthesizing Tight Linear Bounds for Arbitrary Neural Network Activation Functions. It contains code that implements the algorithm described in the paper, scripts to reproduce Tables 1 and 2, and documentation on how to use the scripts. In addition, it also contains a walkthrough for instantiating the LinSyn module to bound an activation function.
 

Files

artifact.zip

Files (3.3 GB)

Name Size Download all
md5:a890ac3df87f13d723b3435b25517860
3.3 GB Preview Download