Bias-Reduced Uncertainty Estimation for Deep Neural Classifiers

TL;DR

This paper introduces a novel uncertainty estimation method for deep neural classifiers that reduces bias by leveraging earlier model snapshots, resulting in more accurate confidence estimates.

Contribution

The paper identifies bias issues in existing uncertainty estimation methods and proposes a new approach using earlier model snapshots to improve accuracy.

Findings

Proposed method outperforms existing uncertainty estimation techniques.

Bias in confidence estimates is linked to training dynamics and overfitting.

Using earlier snapshots reduces bias in high-confidence predictions.

Abstract

We consider the problem of uncertainty estimation in the context of (non-Bayesian) deep neural classification. In this context, all known methods are based on extracting uncertainty signals from a trained network optimized to solve the classification problem at hand. We demonstrate that such techniques tend to introduce biased estimates for instances whose predictions are supposed to be highly confident. We argue that this deficiency is an artifact of the dynamics of training with SGD-like optimizers, and it has some properties similar to overfitting. Based on this observation, we develop an uncertainty estimation algorithm that selectively estimates the uncertainty of highly confident points, using earlier snapshots of the trained model, before their estimates are jittered (and way before they are ready for actual classification). We present extensive experiments indicating that the proposed algorithm provides uncertainty estimates that are consistently better than all known methods.