Skip to content
A library for performing coverage guided fuzzing of neural networks
Branch: master
Clone or download
Augustus Odena
Augustus Odena fix typo
Latest commit 11a6972 Sep 6, 2018
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
bugs First commit Sep 6, 2018
examples First commit Sep 6, 2018
lib
third_party First commit Sep 6, 2018
CONTRIBUTING.md First commit Sep 6, 2018
LICENSE First commit Sep 6, 2018
README.md fix typo Sep 6, 2018
__init__.py First commit Sep 6, 2018
format.sh First commit Sep 6, 2018
requirements.txt First commit Sep 6, 2018

README.md

TensorFuzz: Coverage Guided Fuzzing for Neural Networks

This repository contains a library for performing coverage guided fuzzing of neural networks, as was described in this paper. It's still a prototype, but the ultimate goal is for people to actually use this to test real software. Any suggestions about how to make it more useful for that purpose would be appreciated.

Installation

You ought to be able to run the code in this repository by doing the following:

pip install -r requirements.txt

Then do:

export PYTHONPATH="$PYTHONPATH:$HOME/tensorfuzz"

The structure of this repository

Broadly speaking, this repository contains a core fuzzing library, examples of how to use the fuzzer, a list of bugs found with the fuzzer, and some utilities.

/bugs

This directory contains bugs or weird behaviors that we've found by using this tool.

/examples

This directory contains examples of how to use the fuzzer in several different ways. It contains examples of looking for numerical errors, finding broken loss functions in publicly available code, and checking for disagreements between trained classifiers and their quantized versions.

/lib

This directoy contains the fuzzing engine and all the necessary utils.

/third_party

This directory contains code written by other people and the (potentially updated) LICENSES for that code.

Disclaimers

This is not an officially supported Google product.

You can’t perform that action at this time.