Revisiting Neural Activation Coverage for Uncertainty Estimation

TL;DR

This paper extends neural activation coverage (NAC) to serve as an uncertainty estimation method for trained neural networks in regression tasks, demonstrating its effectiveness over existing techniques like Monte-Carlo Dropout.

Contribution

The authors adapt NAC for uncertainty estimation in regression, showing it provides more meaningful scores than other methods.

Findings

NAC uncertainty scores outperform Monte-Carlo Dropout.

The extended NAC method is effective for trained neural networks.

NAC provides more meaningful uncertainty estimates.

Abstract

Neural activation coverage (NAC) is a recently-proposed technique for out-of-distribution detection and generalization. We build upon this promising foundation and extend the method to work as an uncertainty estimation technique for already-trained artificial neural networks in the domain of regression. Our experiments confirm NAC uncertainty scores to be more meaningful than other techniques, e.g. Monte-Carlo Dropout.