Research map / Label-efficient and robust learning / Semi-supervised learning
The Pessimistic Limits and Possibilities of Margin-based Losses in Semi-supervised Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use a non-identical semi-supervised classifier with a convex decreasing margin loss | Linear classifiers under pessimistic safe SSL · All possible hard labelings of unlabeled data | guaranteed surrogate-loss improvement | Hurt |
| Use soft responsibilities with a convex decreasing margin loss | Linear classifiers under pessimistic safe SSL · All soft assignments of unlabeled data | guaranteed surrogate-loss improvement | Hurt |
| Use unlabeled data with a loss that can increase over some interval, and whose supervised classifier cannot be recovered by any valid responsibilities | Minimax semi-supervised learner · Pessimistic safe SSL | guaranteed surrogate-loss improvement | Helped |
| Use unlabeled points all outside the margin | Convex margin loss decreasing up to margin 1 and increasing beyond it · Pessimistic safe SSL | guaranteed surrogate-loss improvement | Helped |
| Use a quadratic loss and unlabeled data for which no valid responsibility vector recovers the supervised solution | Quadratic-loss classifier · Pessimistic safe SSL | guaranteed surrogate-loss improvement | Helped |
| Use a single unlabeled object outside the interval where absolute-loss responsibilities recover the supervised solution | Absolute-loss classifier · Pessimistic safe SSL | guaranteed surrogate-loss improvement | Helped |
| Rely on unlabeled data for logistic regression without additional assumptions | Logistic regression · Pessimistic safe SSL | guaranteed surrogate-loss improvement | Hurt |
| Rely on unlabeled data for a support vector machine without additional assumptions | Support vector machine · Pessimistic safe SSL | guaranteed surrogate-loss improvement | Hurt |
Closest papers by meaning
- Improved Algorithms for Online Classification with Surrogate Losses (NeurIPS 2026 · similarity 0.89)
- Agnostic Learning of Halfspaces with Gradient Descent via Soft Margins (ICML 2021 · similarity 0.88)
- Optimal Binary Classifier Aggregation for General Losses (NeurIPS 2016 · similarity 0.88)
- On the Error Resistance of Hinge-Loss Minimization (NeurIPS 2020 · similarity 0.87)
- Trading off Consistency and Dimensionality of Convex Surrogates for Multiclass Classification (NeurIPS 2024 · similarity 0.87)
- A Universal Growth Rate for Learning with Smooth Surrogate Losses (NeurIPS 2024 · similarity 0.85)
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.