Research map / Label-efficient and robust learning / Recognition and long-tail
On the Optimization Landscape of Neural Collapse under MSE Loss: Global Optimality with Unconstrained Features
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 rescaled MSE instead of vanilla MSE | ResNet18 · miniImageNet | test performance | Helped |
| Increase either rescaling factor α or M while fixing the other (ablation) | ResNet18 with rescaled MSE · miniImageNet | test performance | Helped |
| Increase feature dimension to at least K−1 (ablation) | ResNet18 with vanilla MSE · CIFAR10 | test accuracy | Helped |
| Train with CE instead of MSE when feature dimension is d=3 (ablation) | ResNet18 · CIFAR10 training features | feature diversity and discriminability | Helped |
| Use a fixed Simplex ETF classifier instead of learning the last-layer classifier (ablation) | ResNet18 with vanilla MSE · CIFAR10 | test accuracy | No clear effect |
| Remove the bias term (ablation) | Unconstrained-feature MSE model · Global optimum | Simplex ETF formation | Hurt |
| Use feature dimension d<K−1 instead of d≥K−1 (ablation) | Unconstrained-feature MSE model · Global optimum | Simplex ETF formation | Hurt |
| Train with SGD, Adam, or L-BFGS (ablation) | ResNet18 with vanilla MSE · CIFAR10 | neural-collapse metrics | No clear effect |
Closest papers by meaning
- Neural Collapse Beyond the Unconstrained Features Model: Landscape, Dynamics, and Generalization in the Mean-Field Regime (ICML 2025 · similarity 0.91)
- The Persistence of Neural Collapse Despite Low-Rank Bias (NeurIPS 2025 · similarity 0.91)
- Neural Collapse is Globally Optimal in Deep Regularized ResNets and Transformers (NeurIPS 2025 · similarity 0.90)
- Neural Collapse under Gradient Flow on Shallow ReLU Networks for Orthogonally Separable Data (NeurIPS 2025 · similarity 0.90)
- Guiding Neural Collapse: Optimising Towards the Nearest Simplex Equiangular Tight Frame (NeurIPS 2024 · similarity 0.90)
- Wide Neural Networks Trained with Weight Decay Provably Exhibit Neural Collapse (ICLR 2025 · similarity 0.90)
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.