Risk-aware Classification via Uncertainty Quantification

TL;DR

This paper enhances deep learning classifiers with uncertainty quantification to enable risk-aware decision-making, crucial for safety-critical applications, by building on Evidential Deep Learning and demonstrating improved empirical performance.

Contribution

It introduces foundational principles for risk-aware classification, extends Evidential Deep Learning to incorporate risk considerations, and empirically validates the improved performance of these methods.

Findings

Superior performance over existing risk-aware classifiers

Effective exercise of discretion in decision-making under uncertainty

Empirical validation of theoretical innovations

Abstract

Autonomous and semi-autonomous systems are using deep learning models to improve decision-making. However, deep classifiers can be overly confident in their incorrect predictions, a major issue especially in safety-critical domains. The present study introduces three foundational desiderata for developing real-world risk-aware classification systems. Expanding upon the previously proposed Evidential Deep Learning (EDL), we demonstrate the unity between these principles and EDL's operational attributes. We then augment EDL empowering autonomous agents to exercise discretion during structured decision-making when uncertainty and risks are inherent. We rigorously examine empirical scenarios to substantiate these theoretical innovations. In contrast to existing risk-aware classifiers, our proposed methodologies consistently exhibit superior performance, underscoring their transformative potential in risk-conscious classification strategies.