Research map / Architectures and efficiency / RNNs and normalization
Neural Stored-program 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 |
|---|---|---|---|
| Add 2 stored programs to NTM | NTM · Associative Recall, test sequences | average bit error per sequence: 7.66 -> 1.35 | Helped |
| Add 2 stored programs to NTM | NTM · Copy, test sequences | average bit error per sequence: 0.00 -> 0.00 | No clear effect |
| Add key-value attention rather than direct attention (ablation) | NUTM on Associative Recall · Associative Recall | training loss convergence | Helped |
| Add the program-key regularization loss (ablation) | NUTM with key-value attention on Associative Recall · Associative Recall | training loss | Helped |
| Use dynamic program selection rather than a fixed uniform program distribution (ablation) | NUTM on Associative Recall · Associative Recall | training loss convergence | Helped |
| Use key-value attention rather than direct attention (ablation) | NUTM on Associative Recall · Associative Recall | training loss convergence | Helped |
| Add a second memory head instead of adding stored programs (ablation) | NTM on Associative Recall · Associative Recall | training loss convergence | Hurt |
| Add 2 stored programs to LRUA | MANN (LRUA) · Omniglot, 5 classes, second presentation, non-persistent memory | classification accuracy: 82.3 -> 85.7 | Helped |
| Add 3 stored programs to LRUA | MANN (LRUA) · Omniglot, 10 classes, second presentation, persistent memory | classification accuracy: 51.3 -> 69.0 | Helped |
| Increase stored programs from 1 to 4 (ablation) | NUTM with DNC core · bAbI, 20 tasks | mean test error (%): 9.7 ± 3.5 -> 5.6 ± 1.9 | Helped |
| Use Neural Stored-program Memory with NTM | NTM · NTM sequencing tasks | bit accuracy: 10-40% | Helped |
| Use Neural Stored-program Memory with NTM | NTM · Continual procedure learning | bit accuracy: 10-40% | Mixed |
Closest papers by meaning
- Neurocoder: General-Purpose Computation Using Stored Neural Programs (ICML 2022 · similarity 0.88)
- Meta-Learning with Memory-Augmented Neural Networks (ICML 2016 · similarity 0.88)
- Blending Complementary Memory Systems in Hybrid Quadratic-Linear Transformers (NeurIPS 2025 · similarity 0.84)
- On the Markov Property of Neural Algorithmic Reasoning: Analyses and Methods (ICLR 2024 · similarity 0.84)
- Continuous Thought Machines (NeurIPS 2025 · similarity 0.84)
- AUTOMEM: Automated Learning of Memory as a Cognitive Skill (NeurIPS 2026 · 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.