HUDD: A tool to debug DNNs for safety analysis

TL;DR

HUDD is a tool that automatically identifies root causes of errors in DNNs through clustering heatmap data and improves DNN safety and accuracy via targeted retraining, aiding safety analysis in critical systems.

Contribution

HUDD introduces a novel heatmap-based clustering method for root cause analysis and an effective retraining approach for DNN safety and accuracy improvement.

Findings

HUDD successfully identifies all distinct root causes of DNN errors.

Retraining with HUDD improves DNN accuracy more effectively than existing methods.

Empirical evaluation in automotive domain demonstrates HUDD's effectiveness.

Abstract

We present HUDD, a tool that supports safety analysis practices for systems enabled by Deep Neural Networks (DNNs) by automatically identifying the root causes for DNN errors and retraining the DNN. HUDD stands for Heatmap-based Unsupervised Debugging of DNNs, it automatically clusters error-inducing images whose results are due to common subsets of DNN neurons. The intent is for the generated clusters to group error-inducing images having common characteristics, that is, having a common root cause. HUDD identifies root causes by applying a clustering algorithm to matrices (i.e., heatmaps) capturing the relevance of every DNN neuron on the DNN outcome. Also, HUDD retrains DNNs with images that are automatically selected based on their relatedness to the identified image clusters. Our empirical evaluation with DNNs from the automotive domain have shown that HUDD automatically identifies all the distinct root causes of DNN errors, thus supporting safety analysis. Also, our retraining approach has shown to be more effective at improving DNN accuracy than existing approaches. A demo video of HUDD is available at https://youtu.be/drjVakP7jdU.