SelectiveNet: A Deep Neural Network with an Integrated Reject Option

TL;DR

SelectiveNet is a deep neural network architecture that jointly learns to classify and reject inputs, improving the trade-off between coverage and accuracy across various datasets.

Contribution

It introduces an end-to-end trainable deep network with an integrated reject option, outperforming existing confidence-based rejection methods.

Findings

Achieves state-of-the-art risk-coverage trade-offs in classification.

Demonstrates improved performance on multiple datasets.

Outperforms traditional confidence threshold methods.

Abstract

We consider the problem of selective prediction (also known as reject option) in deep neural networks, and introduce SelectiveNet, a deep neural architecture with an integrated reject option. Existing rejection mechanisms are based mostly on a threshold over the prediction confidence of a pre-trained network. In contrast, SelectiveNet is trained to optimize both classification (or regression) and rejection simultaneously, end-to-end. The result is a deep neural network that is optimized over the covered domain. In our experiments, we show a consistently improved risk-coverage trade-off over several well-known classification and regression datasets, thus reaching new state-of-the-art results for deep selective classification.