SelectiveNet: A Deep Neural Network with an Integrated Reject Option
arXiv:1901.09192
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.
Accepted to ICML 2019
Cited by in corpus (20)
- WILDS: A Benchmark of in-the-Wild Distribution Shifts
- Self-Adaptive Training: beyond Empirical Risk Minimization
- Adversarial Robustness for Code
- Controlled abstention neural networks for identifying skillful predictions for regression problems
- Controlled abstention neural networks for identifying skillful predictions for classification problems
- DOCTOR: A Simple Method for Detecting Misclassification Errors
- MOS: Towards Scaling Out-of-distribution Detection for Large Semantic Space
- Self-Adaptive Training: Bridging Supervised and Self-Supervised Learning
- Selective Classification Can Magnify Disparities Across Groups
- Beyond Perturbations: Learning Guarantees with Arbitrary Adversarial Test Examples
- Toward Metrics for Differentiating Out-of-Distribution Sets
- ATRO: Adversarial Training with a Rejection Option
- Trading via Selective Classification
- CheXbreak: Misclassification Identification for Deep Learning Models Interpreting Chest X-rays
- On the predictability in reversible steganography
- Dirichlet uncertainty wrappers for actionable algorithm accuracy accountability and auditability
- Selective Classification via One-Sided Prediction
- Online Selective Classification with Limited Feedback
- Improving Compositionality of Neural Networks by Decoding Representations to Inputs
- Selective Regression Under Fairness Criteria