Research map / Learning theory and algorithms / Kernels and regression theory
Neural Collapse Under MSE Loss: Proximity to and Dynamics on the Central Path
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 |
|---|---|---|---|
| Continue MSE training beyond 0-error | VGG, ResNet, and DenseNet · MNIST | test accuracy: 99.23 -> 99.59; 99.16 -> 99.70; 99.62 -> 99.70 | Helped |
| Continue MSE training beyond 0-error | VGG, ResNet, and DenseNet · FashionMNIST | test accuracy: 92.76 -> 92.95; 93.55 -> 93.76; 90.56 -> 92.95 | Helped |
| Continue MSE training beyond 0-error | VGG and ResNet · SVHN | test accuracy: 89.35 -> 93.91; 85.18 -> 92.65 | Helped |
| Continue MSE training beyond 0-error | DenseNet · SVHN | test accuracy: 95.61 -> 95.23 | Hurt |
| Continue MSE training beyond 0-error | VGG, ResNet, and DenseNet · CIFAR10 | test accuracy: 83.13 -> 84.54; 75.43 -> 76.39; 91.77 -> 91.78 | Helped |
| Continue MSE training beyond 0-error | VGG, ResNet, and DenseNet · STL10 | test accuracy: 60.43 -> 67.24; 59.35 -> 60.56; 58.74 -> 60.41 | Helped |
| Train with MSE rather than CE | VGG, ResNet, and DenseNet · MNIST, FashionMNIST, CIFAR10, SVHN, and STL10 | neural collapse | Mixed |
| Train with MSE rather than CE | VGG, ResNet, and DenseNet · MNIST, FashionMNIST, CIFAR10, SVHN, and STL10 | adversarial robustness | Helped |
| Continue training beyond 0-error | MSE-trained networks · Five benchmark datasets | robustness measure: 0.1762 median improvement; 0.9278 mean improvement | Helped |
Closest papers by meaning
- Neural Collapse Beyond the Unconstrained Features Model: Landscape, Dynamics, and Generalization in the Mean-Field Regime (ICML 2025 · similarity 0.90)
- The Persistence of Neural Collapse Despite Low-Rank Bias (NeurIPS 2025 · similarity 0.89)
- Neural Collapse under Gradient Flow on Shallow ReLU Networks for Orthogonally Separable Data (NeurIPS 2025 · similarity 0.89)
- Wide Neural Networks Trained with Weight Decay Provably Exhibit Neural Collapse (ICLR 2025 · similarity 0.88)
- Neural Collapse To Multiple Centers For Imbalanced Data (NeurIPS 2024 · similarity 0.87)
- The Prevalence of Neural Collapse in Neural Multivariate Regression (NeurIPS 2024 · similarity 0.87)
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.