Research map / Learning theory and algorithms / Kernels and regression theory
Understanding Square Loss in Training Overparametrized Neural Network Classifiers
Tianyang Hu, Jun Wang, Wenjia Wang, Zhenguo Li
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 |
|---|---|---|---|
| Replace cross-entropy with square loss and simplex coding | ResNet-18 · CIFAR-10 test | clean accuracy (%): 95.15 (0.11) -> 95.04 (0.07) | No clear effect |
| Replace cross-entropy with square loss and simplex coding | WRN-16-10 · CIFAR-10 test | clean accuracy (%): 93.94 (0.16) -> 95.02 (0.11) | Improved |
| Replace cross-entropy with square loss and simplex coding | ResNet-50 · CIFAR-100 test | clean accuracy (%): 79.82 (0.14) -> 78.91 (0.14) | Hurt |
| Replace cross-entropy with square loss and simplex coding | WRN-16-10 · CIFAR-100 test | clean accuracy (%): 77.89 (0.21) -> 79.65 (0.15) | Improved |
| Replace cross-entropy with square loss and simplex coding | ResNet-18 · CIFAR-10 test, PGD-100, 2/255 | accuracy (%): 8.81 (1.61) -> 30.53 (0.92) | Improved |
| Replace cross-entropy with square loss and simplex coding | WRN-16-10 · CIFAR-10 test, PGD-100, 2/255 | accuracy (%): 1.04 (0.10) -> 37.47 (0.61) | Improved |
| Replace cross-entropy with square loss and simplex coding | ResNet-50 · CIFAR-100 test, PGD-100, 2/255 | accuracy (%): 2.31 (0.07) -> 13.76 (1.30) | Improved |
| Replace cross-entropy with square loss and simplex coding | WRN-16-10 · CIFAR-100 test, PGD-100, 2/255 | accuracy (%): 0.83 (0.07) -> 6.48 (0.40) | Improved |
| Replace cross-entropy with square loss and simplex coding | ResNet-18 · CIFAR-10 test, AutoAttack, 2/255 | accuracy (%): 2.74 (0.09) -> 4.10 (0.50) | Improved |
| Replace cross-entropy with square loss and simplex coding | WRN-16-10 · CIFAR-10 test, AutoAttack, 2/255 | accuracy (%): 0.33 (0.06) -> 5.37 (0.50) | Improved |
| Replace cross-entropy with square loss and simplex coding | ResNet-50 · CIFAR-100 test, AutoAttack, 2/255 | accuracy (%): 0.99 (0.10) -> 3.67 (0.60) | Improved |
| Replace cross-entropy with square loss and simplex coding | WRN-16-10 · CIFAR-100 test, AutoAttack, 2/255 | accuracy (%): 0.42 (0.07) -> 2.73 (0.20) | Improved |
Closest papers by meaning
- Cut your Losses with Squentropy (ICML 2023 · similarity 0.91)
- EVALUATION OF NEURAL ARCHITECTURES TRAINED WITH SQUARE LOSS VS CROSS-ENTROPY IN CLASSIFICATION TASKS (ICLR 2021 · similarity 0.90)
- Are All Losses Created Equal: A Neural Collapse Perspective (NeurIPS 2022 · similarity 0.89)
- Loss Landscape Characterization of Neural Networks without Over-Parametrization (NeurIPS 2024 · similarity 0.87)
- Divergence of Neural Tangent Kernel in Classification Problems (ICLR 2025 · similarity 0.87)
- Multiclass Loss Geometry Matters for Generalization of Gradient Descent in Separable Classification (NeurIPS 2025 · similarity 0.86)
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.
