Research map / Architectures and efficiency / RNNs and normalization
Self-Attentive Associative 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 STM instead of Fast weight | associative retrieval · Associative retrieval, length 50 | test accuracy (%): 20.8 -> 100 | Helped |
| increase item memory dimension from d=48 to d=96 (ablation) | STM · Associative retrieval | epochs to converge: 20 -> 10 (length 30); 80 -> 20 (length 50) | Helped |
| add item-memory gates (ablation) | STM · Associative retrieval | test accuracy (%): 24 -> 100 (length 30); 20 -> 100 (length 50) | Helped |
| add relational transfer (ablation) | STM with d=48 · Associative retrieval, length 50 | epochs to converge: 100 -> 80 | Mixed |
| increase the number of extracted items n_q (ablation) | STM · N-th-farthest | accuracy (%): 84 -> 95 -> 98 | Helped |
| use STM instead of RMC | geometric and graph reasoning · Shortest path | prediction accuracy (%): 66.71 -> 93.43 | Helped |
| use STM instead of MNM-p | bAbI · bAbI | mean error: 0.55 ± 0.74 -> 0.39 ± 0.18 | Helped |
| use STM instead of UT | bAbI · bAbI | mean error: 1.12 ± 1.62 -> 0.39 ± 0.18 | Helped |
| increase model capacity for TPR | TPR · N-th-farthest | convergence | No clear effect |
Closest papers by meaning
- Factual recall in linear associative memories: sharp asymptotics and mechanistic insights (NeurIPS 2026 · similarity 0.86)
- Semantically-correlated memories in a dense associative model (ICML 2024 · similarity 0.86)
- A mathematical theory of balancing relational generalization and memorization (NeurIPS 2026 · similarity 0.86)
- Scaling Laws for Associative Memories (ICLR 2024 · similarity 0.86)
- A Framework for Inference Inspired by Human Memory Mechanisms (ICLR 2024 · similarity 0.85)
- Memory is Reconstructed, Not Retrieved: Graph Memory for LLM Agents (ICML 2026 · 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.