Hierarchical Selective Classification

TL;DR

This paper introduces hierarchical selective classification, a method that leverages class hierarchy to improve uncertainty estimation and control prediction specificity in deep neural networks for risk-sensitive tasks.

Contribution

It formalizes hierarchical risk and coverage, develops algorithms for hierarchical selective classification, and demonstrates improved performance with training regimes like CLIP and knowledge distillation.

Findings

Hierarchical selective classification reduces prediction specificity under uncertainty.

Training regimes like CLIP and knowledge distillation enhance hierarchical selective performance.

The proposed algorithms guarantee target accuracy with high probability.

Abstract

Deploying deep neural networks for risk-sensitive tasks necessitates an uncertainty estimation mechanism. This paper introduces hierarchical selective classification, extending selective classification to a hierarchical setting. Our approach leverages the inherent structure of class relationships, enabling models to reduce the specificity of their predictions when faced with uncertainty. In this paper, we first formalize hierarchical risk and coverage, and introduce hierarchical risk-coverage curves. Next, we develop algorithms for hierarchical selective classification (which we refer to as "inference rules"), and propose an efficient algorithm that guarantees a target accuracy constraint with high probability. Lastly, we conduct extensive empirical studies on over a thousand ImageNet classifiers, revealing that training regimes such as CLIP, pretraining on ImageNet21k and knowledge distillation boost hierarchical selective performance.