Research map / Continual learning / Online and replay-based
Learning to Remember, Learn, and Forget in Attention-Based Models
Djohan Bonnet, Jamie Lohoff, Jan Finkbeiner, Elidona Shiqerukaj, Emre Neftci
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 metaplasticity via fine-tuning (ablation) | Palimpsa-D 760M · Commonsense reasoning average | average accuracy: 51.82 -> 52.27 | Improved |
| add metaplasticity via fine-tuning (ablation) | Palimpsa-M 760M · Commonsense reasoning average | average accuracy: 51.53 -> 52.16 | Improved |
| add metaplasticity via fine-tuning (ablation) | Palimpsa-M 170M · Commonsense reasoning average | average accuracy: 42.90 -> 43.48 | Improved |
| add metaplasticity via fine-tuning (ablation) | Palimpsa-D 170M · Commonsense reasoning average | average accuracy: 43.91 -> 43.96 | No clear effect |
| train metaplasticity from scratch instead of fine-tuning | Palimpsa-D 170M · Commonsense reasoning average | average accuracy: 43.96 -> 43.45 | Hurt |
| add metaplasticity via fine-tuning (ablation) | Palimpsa-M 2.7B · Commonsense reasoning average | average accuracy: 59.22 -> 60.06 | Improved |
| add metaplasticity via fine-tuning (ablation) | Palimpsa-D 760M · LAMBADA | perplexity: 16.54 -> 14.86 | Improved |
| add metaplasticity via fine-tuning (ablation) | Palimpsa-D 170M · WikiText | perplexity: 30.80 -> 29.73 | Improved |
| add metaplasticity via fine-tuning (ablation) | Palimpsa-M 2.7B · WikiText | perplexity: 14.46 -> 14.51 | No clear effect |
| add metaplasticity via fine-tuning (ablation) | Palimpsa-M 2.7B · LongBench overall | accuracy: 19.3 -> 20.4 | Improved |
| add metaplasticity via fine-tuning (ablation) | Palimpsa-M 2.7B · RULER Variable Tracing 4k | accuracy: 59.6 -> 33.5 | Hurt |
| add metaplasticity via fine-tuning (ablation) | Palimpsa-M 2.7B · RULER Variable Tracing 32k | accuracy: 14.9 -> 24.4 | Improved |
Closest papers by meaning
- In-Context Learning as Conditioned Associative Memory Retrieval (ICML 2025 · similarity 0.87)
- Forgetting Transformer: Softmax Attention with a Forget Gate (ICLR 2025 · similarity 0.87)
- The emergence of sparse attention: impact of data distribution and benefits of repetition (NeurIPS 2025 · similarity 0.86)
- Memorization Capacity of Multi-Head Attention in Transformers (ICLR 2024 · similarity 0.86)
- Understanding Factual Recall in Transformers via Associative Memories (ICLR 2025 · similarity 0.86)
- Predictive Attractor Models (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.
