Research map / Optimization / Implicit bias and generalization
Beyond accuracy: generalization properties of bio-plausible temporal credit assignment rules
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 bio-plausible truncated learning rules (e-prop, RFLO, MDGL) instead of BPTT | RNN · sequential MNIST, pattern generation, delayed match-to-sample | generalization gap | Hurt |
| use bio-plausible truncated learning rules instead of BPTT | RNN · three tasks | leading Hessian eigenvalue (curvature) | Hurt |
| scale BPTT update by rho to match three-factor step length along the gradient (ablation) | BPTT · sequential MNIST | curvature convergence | Hurt |
| scale BPTT update by rho to match three-factor step length (ablation) | BPTT · appendix generalization experiment | generalization gap | Hurt |
| increase learning rate of the approximate rule to compensate for small rho | three-factor rule · three tasks | numerical stability | Hurt |
| use a high initial learning rate with gradual decay (ablation) | bio-plausible rules · appendix experiment | curvature of converged solution | Helped |
Closest papers by meaning
- Can Biologically Plausible Temporal Credit Assignment Rules Match BPTT for Neural Similarity? E-prop as an Example (ICML 2025 · similarity 0.89)
- Flatness is Necessary, Neural Collapse is Not: Rethinking Generalization via Grokking (NeurIPS 2025 · similarity 0.88)
- Discovering alternative solutions beyond the simplicity bias in recurrent neural networks (ICLR 2026 · similarity 0.88)
- Do Neural Networks Need Gradient Descent to Generalize? A Theoretical Study (NeurIPS 2025 · similarity 0.87)
- Measuring and Controlling Solution Degeneracy across Task-Trained Recurrent Neural Networks (NeurIPS 2025 · similarity 0.86)
- Counter-Current Learning: A Biologically Plausible Dual Network Approach for Deep Learning (NeurIPS 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.