Classification with Rejection Based on Cost-sensitive Classification
arXiv:2010.11748
Abstract
The goal of classification with rejection is to avoid risky misclassification in error-critical applications such as medical diagnosis and product inspection. In this paper, based on the relationship between classification with rejection and cost-sensitive classification, we propose a novel method of classification with rejection by learning an ensemble of cost-sensitive classifiers, which satisfies all the following properties: (i) it can avoid estimating class-posterior probabilities, resulting in improved classification accuracy, (ii) it allows a flexible choice of losses including non-convex ones, (iii) it does not require complicated modifications when using different losses, (iv) it is applicable to both binary and multiclass cases, and (v) it is theoretically justifiable for any classification-calibrated loss. Experimental results demonstrate the usefulness of our proposed approach in clean-labeled, noisy-labeled, and positive-unlabeled classification.
arXiv comment40 pages. Added the discussion of the recent work by Gangrade et al. (2021) at the end of Section 3.4, where the idea of constructing cost-sensitive classifiers for classification with rejection has also been explored in a different framework of classification with rejection (where the goal is not minimizing the 0-1-c risk as in our paper)
References in corpus (7)
- Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift
- PyTorch: An Imperative Style, High-Performance Deep Learning Library
- Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms
- On Calibration of Modern Neural Networks
- Making Risk Minimization Tolerant to Label Noise
- On Symmetric Losses for Learning from Corrupted Labels
- On the Calibration of Multiclass Classification with Rejection