Research map / Continual learning / Online and replay-based
Continual learning in recurrent neural networks
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 |
|---|---|---|---|
| hypernetwork-based regularization (HNET) instead of weight-importance methods | continual learning of RNNs · Split-AudioSet-10 and Split-SMNIST | final accuracy: AudioSet 71.76 vs 65.56 (Online EWC); SMNIST m=4 94.42 vs 73.16 | Helped |
| increase working memory requirement (pattern manipulation difficulty r from 1 to 5) (ablation) | Online EWC · Pattern Manipulation Task | final accuracy: 95.45 -> 81.80 | Hurt |
| increase working memory requirement (pattern manipulation difficulty r from 1 to 5) (ablation) | HNET · Pattern Manipulation Task | final accuracy: 99.84 -> 93.87 | Hurt |
| increase sequence length with fixed pattern length (Padded/Permuted Copy p=5, i=25) (ablation) | Online EWC and HNET · Permuted Copy Task | final accuracy: Online EWC 97.89 (padded); HNET 100.00 | No clear effect |
| adding Masking+SI (context-dependent gating combined with SI) | weight-importance CL on Split-SMNIST · Split-SMNIST m=1-4 | final accuracy: emerges as preferable method at m=4 (values not in text) | Helped |
| adding Masking+SI on Split-AudioSet-10 | weight-importance CL on AudioSet · Split-AudioSet-10 | final accuracy: 64.86 vs SI 66.92 | Hurt |
| adding generative replay | vanilla RNN · Permuted Copy Task | final accuracy: 100.00 vs Online EWC 98.66 | Helped |
| adding generative replay | LSTM · Split-AudioSet-10 | final accuracy: not successfully trained | No clear effect |
Closest papers by meaning
- Continual learning with hypernetworks (ICLR 2020 · similarity 0.90)
- Revisiting Weight Regularization for Low-Rank Continual Learning (ICLR 2026 · similarity 0.90)
- Theory on Forgetting and Generalization of Continual Learning (ICML 2023 · similarity 0.90)
- TinySubNets: An Efficient and Low Capacity Continual Learning Strategy (AAAI 2025 · similarity 0.88)
- STAR: Stability-Inducing Weight Perturbation for Continual Learning (ICLR 2025 · similarity 0.88)
- The Importance of Being Lazy: Scaling Limits of Continual Learning (ICML 2025 · similarity 0.88)
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.