Research map / Label-efficient and robust learning / Recognition and long-tail
Neural Collapse-Informed Initialization with Perturbation Injection in Classification-based Metric Learning
Jinhee Park, Hee Bin Yoo, Minjun Kim, Byoung-Tak Zhang, Junseok Kwon
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 |
|---|---|---|---|
| add NC-Init to Proxy Anchor (ablation) | Proxy Anchor · CUB (warm-up LR 1e-3) | Recall@1: 81.75 -> 82.70 | Improved |
| add PI on top of NC-Init (ablation) | Proxy Anchor + NC-Init · CUB (warm-up LR 1e-3) | Recall@1: 82.70 -> 83.23 | Improved |
| add NC-Init to Proxy Anchor (ablation) | Proxy Anchor · CARS196 (warm-up LR 5e-3) | Recall@1: 86.80 -> 88.70 | Improved |
| add PI on top of NC-Init (ablation) | Proxy Anchor + NC-Init · CARS196 (warm-up LR 5e-3) | Recall@1: 88.70 -> 89.15 | Improved |
| add NC-Init at smallest warm-up LR 1e-4 (ablation) | Proxy Anchor · CUB (warm-up LR 1e-4) | Recall@1: 81.18 -> 80.33 | Hurt |
| add PI on top of NC-Init at smallest warm-up LR (ablation) | Proxy Anchor + NC-Init · CUB (warm-up LR 1e-4) | Recall@1: 80.33 -> 80.55 | Improved |
| add NC-Init to lower proxy drift (ablation) | Proxy Anchor · CUB (warm-up LR 1e-3) | proxy drift: 2.0146 -> 0.5146 | Improved |
| add NC-Init + PI to Proxy Anchor (ImageNet-21K) | Proxy Anchor · CUB | Recall@1: 81.8 -> 83.2 | Improved |
| add NC-Init + PI to Proxy Anchor (ImageNet-21K) | Proxy Anchor · CARS196 | Recall@1: 86.8 -> 89.2 | Improved |
| add NC-Init + PI to HypViT | HypViT · CARS196 | Recall@1: 86.5 -> 88.5 | Improved |
| add NC-Init + PI to HypViT | HypViT · CUB | Recall@1: 85.6 -> 85.9 | Improved |
| add NC-Init + PI to Proxy Anchor | Proxy Anchor · SOP | Recall@1: 81.8 -> 82.1 | Improved |
Closest papers by meaning
- Guiding Neural Collapse: Optimising Towards the Nearest Simplex Equiangular Tight Frame (NeurIPS 2024 · similarity 0.89)
- Wide Neural Networks Trained with Weight Decay Provably Exhibit Neural Collapse (ICLR 2025 · similarity 0.89)
- Neural Collapse in Cumulative Link Models for Ordinal Regression: An Analysis with Unconstrained Feature Model (NeurIPS 2025 · similarity 0.88)
- Generalized Neural Collapse for a Large Number of Classes (ICML 2024 · similarity 0.87)
- Neural Collapse under Gradient Flow on Shallow ReLU Networks for Orthogonally Separable Data (NeurIPS 2025 · similarity 0.87)
- Neural Collapse Beyond the Unconstrained Features Model: Landscape, Dynamics, and Generalization in the Mean-Field Regime (ICML 2025 · 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.
