Research map / Architectures and efficiency / RNNs and normalization
Contextual Memory Trees
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 |
|---|---|---|---|
| add supervised Update after the initial Insert pass (ablation) | CMT · ALOI | test error: 75.8 -> 26.3 | Helped |
| add supervised Update after the initial Insert pass (ablation) | CMT · WikiPara 1-shot | test error: 97.3 -> 96.7 | Helped |
| increase reroute operations per insertion (ablation) | CMT · WikiPara 1-shot | self-consistency error | Helped |
| increase reroute operations per insertion (ablation) | CMT · WikiPara 1-shot | performance: ~10 | Helped |
| add Update calls over additional training passes (ablation) | CMT · WikiPara 1-shot | training error | Helped |
| increase the leaf-memory multiplier c (ablation) | CMT · ALOI | prediction error | Helped |
| increase the number of examples per label (ablation) | CMT · ALOI S-shot | prediction error | Helped |
| augment OAS with CMT-retrieved memories | OAS · Wiki10-31K | Hamming loss: 20.3 -> 18.3 | Helped |
| augment OAS with CMT-retrieved memories | OAS · RCV1-1K | Hamming loss: 2.6 -> 2.5 | Helped |
| augment OAS with CMT-retrieved memories | OAS · AmazonCat-13K | Hamming loss: 3.0 -> 3.2 | Hurt |
| use CMT instead of nearest-neighbor retrieval | image retrieval · Pascal, Flickr8k, MS COCO | retrieval reward: 10^-3 | No clear effect |
| increase the number of shots | CMT · WikiPara and ImageNet few-shot | test error | Mixed |
Closest papers by meaning
- Learning What to Remember: Test-Time Training via Context Distillation (NeurIPS 2026 · similarity 0.84)
- In-Context Learning as Conditioned Associative Memory Retrieval (ICML 2025 · similarity 0.84)
- Titans: Learning to Memorize at Test Time (NeurIPS 2025 · similarity 0.83)
- Learning from Memory: Non-Parametric Memory Augmented Self-Supervised Learning of Visual Features (ICML 2024 · similarity 0.83)
- CMT: A Memory Compression Method for Continual Knowledge Learning of Large Language Models (AAAI 2025 · similarity 0.83)
- Captured by Captions: On Memorization and its Mitigation in CLIP Models (ICLR 2025 · similarity 0.83)
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.