Failure Prediction by Confidence Estimation of Uncertainty-Aware Dirichlet Networks

TL;DR

This paper introduces a method using uncertainty-aware Dirichlet networks to better predict model failures by estimating confidence, improving safety in critical applications through enhanced error detection.

Contribution

It presents a novel approach for learning true class probabilities that accounts for imbalance and TCP constraints, improving failure prediction in deep learning models.

Findings

Enhanced separation between correct and incorrect prediction confidence.

Improved failure prediction over baseline methods.

Effective across multiple image classification architectures.

Abstract

Reliably assessing model confidence in deep learning and predicting errors likely to be made are key elements in providing safety for model deployment, in particular for applications with dire consequences. In this paper, it is first shown that uncertainty-aware deep Dirichlet neural networks provide an improved separation between the confidence of correct and incorrect predictions in the true class probability (TCP) metric. Second, as the true class is unknown at test time, a new criterion is proposed for learning the true class probability by matching prediction confidence scores while taking imbalance and TCP constraints into account for correct predictions and failures. Experimental results show our method improves upon the maximum class probability (MCP) baseline and predicted TCP for standard networks on several image classification tasks with various network architectures.