Skip to content

Latest commit

 

History

History
8 lines (6 loc) · 563 Bytes

README.md

File metadata and controls

8 lines (6 loc) · 563 Bytes

Tensorflow Ops for Deep Neural Network Attribution.

Uses the Integrated Gradients Technique from the paper: https://arxiv.org/abs/1703.01365

This directory contains functions for augmenting an existing tensorflow graph in-place to add ops for computing Integrated Gradients. It uses the Tensorflow Graph Editor to add ops that perform the entire Integrated Gradient computation within a single sess.run call. It works for both continuous inputs, like images, and discrete inputs with embeddings. Example usage of the functions can be found in the test cases.