Risk-aware Classification via Uncertainty Quantification
This addresses safety concerns in autonomous systems by improving uncertainty quantification for risk-aware classification, though it appears incremental as it builds on existing EDL methods.
The paper tackles the problem of deep classifiers being overly confident in incorrect predictions in safety-critical domains by introducing three foundational desiderata for risk-aware classification systems and augmenting Evidential Deep Learning (EDL) to enable autonomous agents to exercise discretion during decision-making. The proposed methodologies consistently exhibit superior performance compared to existing risk-aware classifiers.
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.