Research map / Optimization / Implicit bias and generalization
A Theoretical Analysis of the Learning Dynamics under Class Imbalance
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 per-class-normalized GD instead of GD | GD · Bi60, Mod1 | minority-class recall | Helped |
| Use per-class-normalized GD instead of GD | GD · Imbalanced classification; multiple architectures and datasets | training and test performance | Helped |
| Use per-class-normalized SGD without balancing directional noise (ablation) | SGD · Imbalanced classification | minority-class recall | No clear effect |
| Rescale the minority-class signal to compensate for directional noise (ablation) | PCNSGD · Bi7a, Mod1 | minority-class recall | Helped |
| Use equal numbers of examples from each class in every minibatch | SGD · Imbalanced classification | minority-class recall | Helped |
| Combine per-class normalization with oversampling (ablation) | SGD · Bi7a, Mod1 | test recall | Mixed |
| Use oversampling with PCNSGD instead of minority-signal rescaling (ablation) | PCNSGD+R · Bi7a, Mod1 | recall | Helped |
| Apply GD to an imbalanced dataset rather than a balanced one | GD · Imbalanced classification | minority-class loss | Hurt |
| Use PCN with full-batch GD to equalize per-class gradient magnitudes (ablation) | GD · Imbalanced classification | per-class loss monotonicity | Helped |
| Use PCN with SGD without increasing batch size or balancing noise (ablation) | SGD · Imbalanced classification | per-class loss monotonicity | Hurt |
Closest papers by meaning
- Balancing the Scales: A Theoretical and Algorithmic Framework for Learning from Imbalanced Data (ICML 2025 · similarity 0.90)
- Restoring balance: principled under/oversampling of data for optimal classification (ICML 2024 · similarity 0.89)
- Learning under Localized Minority Imbalance (NeurIPS 2026 · similarity 0.88)
- Balanced Data, Imbalanced Spectra: Unveiling Class Disparities with Spectral Imbalance (ICML 2024 · similarity 0.88)
- Imbalances in Neurosymbolic Learning: Characterization and Mitigating Strategies (NeurIPS 2025 · similarity 0.87)
- Gradient Reweighting: Towards Imbalanced Class-Incremental Learning (CVPR 2024 · 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.