Research map / Label-efficient and robust learning / Recognition and long-tail
Deep Gamblers: Learning to Abstain with Portfolio Theory
Ziyin Liu, Zhikang Wang, Paul Pu Liang, Ruslan Salakhutdinov, Louis-Philippe Morency, Masahito Ueda
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Train with the gambler's loss and an added reservation output instead of standard cross-entropy | VGG16 classifier · SVHN, coverage 0.95 | error percentage: 1.39 -> 1.36 | Improved |
| Train with the gambler's loss and an added reservation output instead of standard cross-entropy | VGG16 classifier · CIFAR10, coverage 0.95 | error percentage: 4.16 -> 3.49 ± 0.15 | Improved |
| Train with the gambler's loss and an added reservation output instead of standard cross-entropy | VGG16 classifier · Cats vs. Dogs, coverage 0.95 | error percentage: 1.62 -> 1.23 ± 0.12 | Improved |
| Train with the gambler's loss and an added reservation output instead of standard cross-entropy | VGG16 classifier · SVHN, coverage 1.00 | error percentage: 3.21 -> 3.24 ± 0.09 | Hurt |
| Use the best per-coverage hyperparameter setting rather than the best single model | Deep Gamblers · SVHN, coverage 0.85 | error percentage: 0.57 ± 0.07 -> 0.66 ± 0.01 | Hurt |
| Use the best per-coverage hyperparameter setting rather than the best single model | Deep Gamblers · Cats vs. Dogs, coverage 0.95 | error percentage: 1.23 ± 0.12 -> 0.88 ± 0.38 | Improved |
| Use a lower reward parameter o (ablation) | Deep Gamblers · CIFAR10 | training convergence: null | Hurt |
| Pretrain with cross-entropy before switching to the gambler's loss for low o (ablation) | Deep Gamblers · CIFAR10 | prediction accuracy: null | No clear effect |
| Train with a lower o (ablation) | Deep Gamblers · various datasets and coverages | error at low coverage: null | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SelectiveNet — SelectiveNet: A Deep Neural Network with an Integrated Reject Option (ICML 2019) · better on 10 of 13 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Learning Not to Regret (AAAI 2024 · similarity 0.86)
- Coin Betting and Parameter-Free Online Learning (NeurIPS 2016 · similarity 0.85)
- Online Learning in Betting Markets: Profit versus Prediction (ICML 2024 · similarity 0.85)
- Deep Learning Games (NeurIPS 2016 · similarity 0.84)
- Weisfeiler and Leman Go Gambling: Why Expressive Lottery Tickets Win (ICML 2025 · similarity 0.84)
- A representation-learning game for classes of prediction tasks (ICLR 2024 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
