Research map / Architectures and efficiency / RNNs and normalization
Metalearned Neural Memory
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 learned local memory updates instead of gradient-based updates (ablation) | MNM · bAbI word-level | Mean Error: 3.2 $\pm$ 0.5 -> 0.55 $\pm$0.74 | Helped |
| Use learned local memory updates instead of gradient-based updates (ablation) | MNM · bAbI word-level | Failed Tasks (>5% error): 1.3 $\pm$ 0.8 -> 0.08 $\pm$ 0.28 | Helped |
| Use learned local updates rather than gradient-based memory updates (ablation) | MNM · Dictionary inference, harder task instances | Task error | Mixed |
| Use learned local updates rather than gradient-based memory updates (ablation) | MNM · Algorithmic tasks | Wallclock time | Helped |
| Use a neural memory with a learned memory access mechanism instead of a soft-attention lookup table | Word-level bAbI model · bAbI word-level | Mean Error: 77.5% -> 9.5% | Helped |
| Use MNM instead of differential-plasticity agents | Maze exploration agent · 9 × 9 maze | Training episodes to reach reward 175: nearly 1M -> 250K | Helped |
| Use MNM instead of differential-plasticity agents | Maze exploration agent · 9 × 9 maze after 1M episodes | Reward | Helped |
| Use a neural memory with learned local updates instead of the soft-attention lookup table | Algorithmic task model · Priority sort | Task performance | Hurt |
Closest papers by meaning
- Tell Me What To Learn: Generalizing Neural Memory to be Controllable in Natural Language (NeurIPS 2026 · similarity 0.86)
- Titans: Learning to Memorize at Test Time (NeurIPS 2025 · similarity 0.85)
- AUTOMEM: Automated Learning of Memory as a Cognitive Skill (NeurIPS 2026 · similarity 0.85)
- MemReward: Graph-Based Experience Memory for LLM Reward Prediction with Limited Labels (NeurIPS 2026 · similarity 0.84)
- Blending Complementary Memory Systems in Hybrid Quadratic-Linear Transformers (NeurIPS 2025 · similarity 0.84)
- From RAG to Memory: Non-Parametric Continual Learning for Large Language Models (ICML 2025 · similarity 0.84)
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.