Research map / Learning theory and algorithms / Prediction and decision losses
Learning with Average Top-k Loss
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 |
|---|---|---|---|
| Replace average loss with AT_k* loss | Logistic-loss classifier · UCI/KEEL binary classification; selected examples: Monk, Australian, Madelon, Splice, German, Phoneme | misclassification rate (%): 20.46(2.02) -> 16.76(2.29) on Monk; 14.27(3.22) -> 11.70(2.82) on Australian | Helped |
| Replace average loss with AT_k* loss | Hinge-loss classifier · UCI/KEEL binary classification; selected examples: Monk, Australian, Titanic | misclassification rate (%): 18.61(3.16) -> 17.04(2.77) on Monk; 14.74(3.10) -> 12.51(4.03) on Australian | Helped |
| Replace average loss with AT_k* loss | Regression with square loss · Sinc, Housing, Abalone, Cpusmall | RMSE: 0.1147(0.0060) -> 0.1139(0.0057) on Sinc; 0.1065(0.0132) -> 0.1050(0.0132) on Housing | Helped |
| Use AT_k* instead of average loss | Logistic-loss classifier · Spambase | misclassification rate (%): 8.36(0.97) -> 8.36(0.97) | No clear effect |
| Use AT_k* instead of average loss | Hinge-loss classifier · Spambase and Phoneme | misclassification rate (%): 7.40(0.72) -> 7.40(0.72) on Spambase; 22.88(1.01) -> 22.88(1.01) on Phoneme | No clear effect |
| Use AT_k* instead of average loss | Hinge-loss classifier · Splice | misclassification rate (%): 16.25(1.12) -> 16.23(0.97) | No clear effect |
| Use an intermediate k rather than k=1 (maximum loss) or k=n (average loss) (ablation) | MAT_k classifier · Four representative binary-classification datasets | misclassification rate | Helped |
| Increase k from 1 to intermediate values (ablation) | MAT_k classifier · Monk, Australian, Splice | misclassification rate | Helped |
| Increase k toward n, including k=n (ablation) | MAT_k classifier · Monk, Australian, Phoneme | misclassification rate | Hurt |
Closest papers by meaning
- Loss Functions for Top-k Error: Analysis and Insights (CVPR 2016 · similarity 0.89)
- Stochastic smoothing of the top-K calibrated hinge loss for deep imbalanced classification (ICML 2022 · similarity 0.88)
- Top-k Hierarchical Classification (AAAI 2017 · similarity 0.88)
- EnsLoss: Stochastic Calibrated Loss Ensembles for Preventing Overfitting in Classification (ICML 2025 · similarity 0.85)
- A Survey of Learning Criteria Going beyond the Usual Risk (Abstract Reprint) (AAAI 2024 · similarity 0.85)
- Multiclass Loss Geometry Matters for Generalization of Gradient Descent in Separable Classification (NeurIPS 2025 · similarity 0.84)
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.