Black-box Safety Analysis and Retraining of DNNs based on Feature Extraction and Clustering

TL;DR

This paper introduces SAFE, a black-box method that uses feature extraction and clustering to identify error causes in DNNs and improve their accuracy without internal access, especially in safety-critical applications.

Contribution

SAFE is a novel black-box approach that automatically characterizes root causes of DNN errors using transfer learning and clustering, enabling effective retraining and safety analysis.

Findings

SAFE effectively identifies diverse root causes of errors.

SAFE improves DNN accuracy significantly after retraining.

SAFE reduces execution time and memory compared to alternatives.

Abstract

Deep neural networks (DNNs) have demonstrated superior performance over classical machine learning to support many features in safety-critical systems. Although DNNs are now widely used in such systems (e.g., self driving cars), there is limited progress regarding automated support for functional safety analysis in DNN-based systems. For example, the identification of root causes of errors, to enable both risk analysis and DNN retraining, remains an open problem. In this paper, we propose SAFE, a black-box approach to automatically characterize the root causes of DNN errors. SAFE relies on a transfer learning model pre-trained on ImageNet to extract the features from error-inducing images. It then applies a density-based clustering algorithm to detect arbitrary shaped clusters of images modeling plausible causes of error. Last, clusters are used to effectively retrain and improve the DNN. The black-box nature of SAFE is motivated by our objective not to require changes or even access to the DNN internals to facilitate adoption. Experimental results show the superior ability of SAFE in identifying different root causes of DNN errors based on case studies in the automotive domain. It also yields significant improvements in DNN accuracy after retraining, while saving significant execution time and memory when compared to alternatives.