Research map / Learning theory and algorithms / Prediction and decision losses
Cross-Entropy Loss Functions: Theoretical Analysis and Applications
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 adv-comp-sum instead of TRADES | WRN-70-16 · CIFAR-10 | clean accuracy: 85.34 ± 0.04 -> 86.16 ± 0.16 | Helped |
| Use adv-comp-sum instead of TRADES | WRN-70-16 · CIFAR-10 | PGD40 margin accuracy: 57.90 ± 0.13 -> 59.35 ± 0.07 | Helped |
| Use adv-comp-sum instead of TRADES | WRN-70-16 · CIFAR-10 | AutoAttack accuracy: 57.05 ± 0.17 -> 57.77 ± 0.08 | Helped |
| Use adv-comp-sum instead of TRADES | WRN-34-20 and WRN-28-10 · CIFAR-10 | PGD40 margin accuracy: 57.54 ± 0.18 -> 58.92 ± 0.06; 55.92 ± 0.20 -> 57.28 ± 0.05 | Helped |
| Use adv-comp-sum instead of TRADES | WRN-34-20 and WRN-28-10 · CIFAR-10 | clean accuracy: 85.21 ± 0.16 -> 85.59 ± 0.17; 84.33 ± 0.18 -> 84.50 ± 0.33 | Helped |
| Use adv-comp-sum instead of TRADES | WRN-34-20 and WRN-28-10 · CIFAR-10 | AutoAttack accuracy: 56.70 ± 0.14 -> 57.41 ± 0.06; 55.19 ± 0.23 -> 55.79 ± 0.06 | Helped |
| Use adv-comp-sum instead of TRADES | WRN-70-16 · CIFAR-100 | clean accuracy: 60.56 ± 0.31 -> 63.10 ± 0.24 | Helped |
| Use adv-comp-sum instead of TRADES | WRN-70-16 · CIFAR-100 | AutoAttack accuracy: 29.93 ± 0.14 -> 31.05 ± 0.15 | Helped |
| Use adv-comp-sum instead of TRADES | WRN-34-20 · SVHN | clean accuracy: 93.03 ± 0.13 -> 93.98 ± 0.12 | Helped |
| Use adv-comp-sum instead of TRADES | WRN-34-20 · SVHN | AutoAttack accuracy: 57.84 ± 0.19 -> 58.13 ± 0.12 | Helped |
| Increase comp-sum τ from 0 to 1 | Comp-sum loss · CIFAR-10 | accuracy: 87.37 ± 0.57 -> 92.59 ± 0.10 | Helped |
| Increase comp-sum τ from 0 to 1 | Comp-sum loss · CIFAR-100 | accuracy: 57.87 ± 0.60 -> 70.93 ± 0.34 | Helped |
Closest papers by meaning
- Rethinking Softmax Cross-Entropy Loss for Adversarial Robustness (ICLR 2020 · similarity 0.89)
- Are All Losses Created Equal: A Neural Collapse Perspective (NeurIPS 2022 · similarity 0.89)
- Expressive Losses for Verified Robustness via Convex Combinations (ICLR 2024 · similarity 0.86)
- Diagonalizing the Softmax: Hadamard Initialization for Tractable Cross-Entropy Dynamics (NeurIPS 2026 · similarity 0.86)
- Loss Functions and Operators Generated by f-Divergences (ICML 2025 · similarity 0.86)
- Decoupled Kullback-Leibler Divergence Loss (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.