Hybrid computing using a neural network with dynamic external memory
Abstract
Artificial neural networks are remarkably adept at sensory processing, sequence learning and reinforcement learning, but are limited in their ability to represent variables and data structures and to store data over long timescales, owing to the lack of an external memory. Here we introduce a machine learning model called a differentiable neural computer (DNC), which consists of a neural network that can read from and write to an external memory matrix, analogous to the random-access memory in a conventional computer. Like a conventional computer, it can use its memory to represent and manipulate complex data structures, but, like a neural network, it can learn to do so from data. When trained with supervised learning, we demonstrate that a DNC can successfully answer synthetic questions designed to emulate reasoning and inference problems in natural language. We show that it can learn tasks such as finding the shortest path between specified points and inferring the missing links in randomly generated graphs, and then generalize these tasks to specific graphs such as transport networks and family trees. When trained with reinforcement learning, a DNC can complete a moving blocks puzzle in which changing goals are specified by sequences of symbols. Taken together, our results demonstrate that DNCs have the capacity to solve complex, structured tasks that are inaccessible to neural networks without external read-write memory.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
Differentiable Neural Computers suffer from poor computational scaling.
“Differentiable Neural Computers offer differentiable memory operations but suffer from poor computational scaling.”
From ARTEM: Enhancing Large Language Model Agents with Spatial-Temporal Episodic Memory (2026) · cites this paper as Graves et al. 2016
Memory-augmented neural networks (MemN2N, NTM, DNC) focus on algorithmic tasks rather than long-term personalization.
“MemN2N [20], NTM [21], and DNC [22] introduced differentiable external memory for neural networks. However, these focus on algorithmic tasks rather than long-term personalization.”
From From Goldfish to Elephant: Long-Term Memory for Large Language Models via Selective Storage and Hierarchical Compression (2026) · cites this paper as [22]
DNC enhanced memory addressing and management algorithms by applying structured reasoning tasks using graphs.
“Building on this, Graves et al. (2016) proposed the Differentiable Neural Computer (DNC), which enhanced memory addressing and management algorithms by applying structured reasoning tasks using graphs.”
From Memory Fabric for Conversational AI Agents: Enabling Shared and Persistent Memory Across Users (2025) · cites this paper as Graves et al. (2016)
DNCs demonstrated the ability to store and traverse graph-like structures using learned addressing.
“DNCs demonstrated the ability to store and traverse graph-like structures using learned addressing.”
From Memory Fabric for Conversational AI Agents: Enabling Shared and Persistent Memory Across Users (2025) · cites this paper as Graves et al. (2016)
A hybrid memory approach combined with short-term memory, such as attention layers within transformer-based models, along with longterm memory storage using vector databases or key-value memory structures, allows conversational agents to maintain continuity across sessions while efficiently retrieving relevant past interactions or external knowledge.
“This combination allows conversational agents to maintain continuity across sessions while efficiently retrieving relevant past interactions or external knowledge.Thereby providing enhanced personalization and contextual awareness [12,13,28].”
From Memory Fabric for Conversational AI Agents: Enabling Shared and Persistent Memory Across Users (2025) · cites this paper as [12]
Architectures for memory-like functions have led to meaningful improvements in capabilities.
“Architectures for incorporating different kinds of memorylike functions have led in the past to meaningful improvements in capabilities.”
From Episodic Memory in AI Agents Poses Risks that Should be Studied and Mitigated (2025) · cites this paper as [30]
Cites
ImageNet classification with deep convolutional neural networks
Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2017) · Communications of the ACM · doi:10.1145/3065386 · Cited by 112,155 · see citing works
Hochreiter, S., & Schmidhuber, J. (1997) · Neural Computation · doi:10.1162/neco.1997.9.8.1735 · Cited by 102,318 · see citing works
Bengio, Y., Louradour, J., Collobert, R., & Weston, J. (2009) · Proceedings of the 26th Annual International Conference on Machine Learning · doi:10.1145/1553374.1553380 · Cited by 5,366 · cited 1× · see citing works
“After training with curriculum learning 28,29 using graphs and queries with gradually increasing complexity, the networks were tested (with no retraining) on two specific graphs as a test of generalization to realistic data: a symbolic map of the London Underground and an invented family tree.”
MethodMcClelland, J. L., McNaughton, B. L., & O'Reilly, R. C. (1995) · Psychological Review · doi:10.1037/0033-295x.102.3.419 · Cited by 5,256 · see citing works
Human-level concept learning through probabilistic program induction
Lake, B. M., Salakhutdinov, R., & Tenenbaum, J. B. (2015) · Science · doi:10.1126/science.aab3050 · Cited by 2,817 · see citing works
The role of context in object recognition
Oliva, A., & Torralba, A. (2007) · Trends in Cognitive Sciences · doi:10.1016/j.tics.2007.09.009 · Cited by 1,063 · see citing works
What Learning Systems do Intelligent Agents Need? Complementary Learning Systems Theory Updated
Kumaran, D., Hassabis, D., & McClelland, J. L. (2016) · Trends in Cognitive Sciences · doi:10.1016/j.tics.2016.05.004 · Cited by 736 · see citing works
Self-citationThe Algebraic Mind: Integrating Connectionism and Cognitive Science
Marcus, G. F. (2001) · doi:10.7551/mitpress/1187.001.0001 · Cited by 341 · see citing works
Flagged first, then by citations
Cited by
Sorted by citations
Continual lifelong learning with neural networks: A review
Parisi, G. I., Kemker, R., Part, J. L., Kanan, C., & Wermter, S. (2019) · Neural Networks · doi:10.1016/j.neunet.2019.01.012 · Cited by 3,117 · see citing works
An Introduction to Deep Learning for the Physical Layer
O'Shea, T., & Hoydis, J. (2017) · IEEE Transactions on Cognitive Communications and Networking · doi:10.1109/tccn.2017.2758370 · Cited by 2,980 · see citing works
On the Opportunities and Risks of Foundation Models
Bommasani, R., Hudson, D. A., Adeli, E., Altman, R., Arora, S., von Arx, S., Bernstein, M. S., Bohg, J., Bosselut, A., Brunskill, E., et al. (2021) · arXiv · doi:10.48550/arxiv.2108.07258 · Cited by 2,676 · see citing works
Building machines that learn and think like people
Lake, B. M., Ullman, T. D., Tenenbaum, J. B., & Gershman, S. J. (2017) · Behavioral and Brain Sciences · doi:10.1017/s0140525x16001837 · Cited by 2,550 · see citing works
Memory devices and applications for in-memory computing
Sebastian, A., Le Gallo, M., Khaddam-Aljameh, R., & Eleftheriou, E. (2020) · Nature Nanotechnology · doi:10.1038/s41565-020-0655-z · Cited by 2,289 · see citing works
The rise of deep learning in drug discovery
Chen, H., Engkvist, O., Wang, Y., Olivecrona, M., & Blaschke, T. (2018) · Drug Discovery Today · doi:10.1016/j.drudis.2018.01.039 · Cited by 1,770 · see citing works
Neuroscience-Inspired Artificial Intelligence
Hassabis, D., Kumaran, D., Summerfield, C., & Botvinick, M. (2017) · Neuron · doi:10.1016/j.neuron.2017.06.011 · Cited by 1,557 · see citing works
Self-citationEvaluating Large Language Models Trained on Code
Chen, M., Tworek, J., Jun, H., Yuan, Q., Pinto, H. P. D. O., Kaplan, J., Edwards, H., Burda, Y., Joseph, N., Brockman, G., et al. (2021) · arXiv · doi:10.48550/arxiv.2107.03374 · Cited by 1,436 · see citing works
“…incorporated inductive biases based on modern computing devices, such as the Neural Turing Machine (Graves et al., 2014), memory networks (Weston et al., 2015; Sukhbaatar et al., 2015), the Neural GPU (Kaiser & Sutskever, 2015), and the differentiable neural computer (Graves et al., 2016).”
BackgroundMethodA clinically applicable approach to continuous prediction of future acute kidney injury
Tomašev, N., Glorot, X., Rae, J. W., Zielinski, M., Askham, H., Saraiva, A., Mottram, A., Meyer, C., Ravuri, S., Protsyuk, I., et al. (2019) · Nature · doi:10.1038/s41586-019-1390-1 · Cited by 1,086 · see citing works
Self-citationDeep Learning: A Critical Appraisal
Marcus, G. (2018) · arXiv · doi:10.48550/arxiv.1801.00631 · Cited by 853 · see citing works
Deep Reinforcement Learning: An Overview
Li, Y. (2017) · arXiv · doi:10.48550/arxiv.1701.07274 · Cited by 762 · see citing works
Deep learning for molecular design—a review of the state of the art
Elton, D. C., Boukouvalas, Z., Fuge, M. D., & Chung, P. W. (2019) · Molecular Systems Design & Engineering · doi:10.1039/c9me00039a · Cited by 683 · see citing works
Reinforcement learning, fast and slow
Botvinick, M., Ritter, S., Wang, J. X., Kurth-Nelson, Z., Blundell, C., & Hassabis, D. (2019) · Trends in Cognitive Sciences · doi:10.1016/j.tics.2019.02.006 · Cited by 667 · see citing works
Self-citationLearning to reinforcement learn
Wang, J. X., Kurth-Nelson, Z., Tirumala, D., Soyer, H., Leibo, J. Z., Munos, R., Blundell, C., Kumaran, D., & Botvinick, M. (2016) · arXiv · doi:10.48550/arxiv.1611.05763 · Cited by 579 · see citing works
Beyond the Imitation Game: Quantifying and extrapolating the capabilities of language models
Srivastava, A., Rastogi, A., Rao, A., Shoeb, A. A. M., Abid, A., Fisch, A., Brown, A. R., Santoro, A., Gupta, A., Garriga-Alonso, A., et al. (2022) · arXiv · doi:10.48550/arxiv.2206.04615 · Cited by 569 · see citing works
If deep learning is the answer, what is the question?
Saxe, A., Nelli, S., & Summerfield, C. (2021) · Nature Reviews Neuroscience · doi:10.1038/s41583-020-00395-8 · Cited by 397 · see citing works
Self-citationProgram Synthesis with Large Language Models
Austin, J., Odena, A., Nye, M., Bosma, M., Michalewski, H., Dohan, D., Jiang, E., Cai, C., Terry, M., Le, Q., et al. (2021) · arXiv · doi:10.48550/arxiv.2108.07732 · Cited by 388 · see citing works
“…are deep network architectures that aim to learn algorithms from input-output examples, by structuring the network in a way that corresponds to mathematical models of computation like Turing machines [Graves et al., 2014, Kaiser and Sutskever, 2016, Kurach et al., 2016, Graves et al., 2016].”
Direct Fit to Nature: An Evolutionary Perspective on Biological and Artificial Neural Networks
Hasson, U., Nastase, S. A., & Goldstein, A. (2020) · Neuron · doi:10.1016/j.neuron.2019.12.002 · Cited by 387 · see citing works
GAMENet: Graph Augmented MEmory Networks for Recommending Medication Combination
Shang, J., Xiao, C., Ma, T., Li, H., & Sun, J. (2019) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v33i01.33011126 · Cited by 345 · see citing works
Program Induction by Rationale Generation: Learning to Solve and Explain Algebraic Word Problems
Ling, W., Yogatama, D., Dyer, C., & Blunsom, P. (2017) · Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/p17-1015 · Cited by 325 · see citing works
Self-citationDeep reinforcement learning and its neuroscientific implications
Botvinick, M., Wang, J. X., Dabney, W., Miller, K. J., & Kurth-Nelson, Z. (2020) · Neuron · doi:10.1016/j.neuron.2020.06.014 · Cited by 269 · see citing works
Camps, J., Samà, A., Martín, M., Rodríguez-Martín, D., Pérez-López, C., Moreno Arostegui, J. M., Cabestany, J., Català, A., Alcaine, S., Mestre, B., et al. (2018) · Knowledge-Based Systems · doi:10.1016/j.knosys.2017.10.017 · Cited by 224 · see citing works
Compressive Transformers for Long-Range Sequence Modelling
Rae, J. W., Potapenko, A., Jayakumar, S. M., & Lillicrap, T. P. (2019) · arXiv · doi:10.48550/arxiv.1911.05507 · Cited by 205 · see citing works
Radiology report generation with a learned knowledge base and multi-modal alignment
Yang, S., Wu, X., Ge, S., Zheng, Z., Zhou, S. K., & Xiao, L. (2023) · Medical Image Analysis · doi:10.1016/j.media.2023.102798 · Cited by 199 · see citing works
Koster, R., Chadwick, M. J., Chen, Y., Berron, D., Banino, A., Düzel, E., Hassabis, D., & Kumaran, D. (2018) · Neuron · doi:10.1016/j.neuron.2018.08.009 · Cited by 191 · see citing works
Scene Memory Transformer for Embodied Agents in Long-Horizon Tasks
Fang, K., Toshev, A., Fei-Fei, L., & Savarese, S. (2019) · 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2019.00063 · Cited by 178 · see citing works
Neural Map: Structured Memory for Deep Reinforcement Learning
Parisotto, E., & Salakhutdinov, R. (2017) · arXiv · doi:10.48550/arxiv.1702.08360 · Cited by 178 · see citing works
The Rhetoric and Reality of Anthropomorphism in Artificial Intelligence
Watson, D. (2019) · Minds and Machines · doi:10.1007/s11023-019-09506-6 · Cited by 148 · see citing works
Deep learning and the global workspace theory
VanRullen, R., & Kanai, R. (2021) · Trends in Neurosciences · doi:10.1016/j.tins.2021.04.005 · Cited by 128 · see citing works
Rethinking search: making domain experts out of dilettantes
Metzler, D., Tay, Y., Bahri, D., & Najork, M. (2021) · ACM SIGIR Forum · doi:10.1145/3476415.3476428 · Cited by 112 · see citing works
Ueltzhöffer, K. (2018) · Biological Cybernetics · doi:10.1007/s00422-018-0785-7 · Cited by 98 · see citing works
McClelland, J. L., Hill, F., Rudolph, M., Baldridge, J., & Schütze, H. (2020) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.1910416117 · Cited by 91 · see citing works
Symbol Emergence in Cognitive Developmental Systems: A Survey
Taniguchi, T., Piater, J., Worgotter, F., Ugur, E., Hoffmann, M., Jamone, L., Nagai, T., Rosman, B., Matsuka, T., Iwahashi, N., et al. (2019) · IEEE Transactions on Cognitive and Developmental Systems · doi:10.1109/tcds.2018.2867772 · Cited by 92 · see citing works
Clune, J. (2019) · arXiv · doi:10.48550/arxiv.1905.10985 · Cited by 82 · see citing works
Efficient probabilistic inference in generic neural networks trained with non-probabilistic feedback
Orhan, A. E., & Ma, W. J. (2017) · Nature Communications · doi:10.1038/s41467-017-00181-8 · Cited by 77 · see citing works
Bulatov, A., Kuratov, Y., & Burtsev, M. S. (2022) · arXiv · doi:10.48550/arxiv.2207.06881 · Cited by 55 · see citing works
Casares, A. P. (2018) · Futures · doi:10.1016/j.futures.2018.05.002 · Cited by 54 · see citing works
CLEVR: A Diagnostic Dataset for Compositional Language and Elementary Visual Reasoning
Johnson, J., Hariharan, B., van der Maaten, L., Fei-Fei, L., Zitnick, C. L., & Girshick, R. (2016) · arXiv · doi:10.48550/arxiv.1612.06890 · Cited by 40 · see citing works
SAGE: Self-evolving Agents with Reflective and Memory-augmented Abilities
Liang, X., Tao, M., Xia, Y., Wang, J., Li, K., Wang, Y., He, Y., Yang, J., Shi, T., Wang, Y., et al. (2025) · Neurocomputing · doi:10.1016/j.neucom.2025.130470 · Cited by 39 · see citing works
Testolin, A. (2024) · Applied Sciences · doi:10.3390/app14020744 · Cited by 24 · see citing works
Show Your Work: Scratchpads for Intermediate Computation with Language Models
Nye, M., Andreassen, A. J., Gur-Ari, G., Michalewski, H., Austin, J., Bieber, D., Dohan, D., Lewkowycz, A., Bosma, M., Luan, D., et al. (2021) · arXiv · doi:10.48550/arxiv.2112.00114 · Cited by 144 · see citing works
Zheng, S., He, K., Yang, L., & Xiong, J. (2024) · IEEE Access · doi:10.1109/access.2024.3393485 · Cited by 18 · see citing works
“A significant development in this domain is the progress in Memory-Augmented Neural Networks (MANNs), which address traditional neural networks’ limitations in handling complex mappings and relearning from fresh data [21], [22].”
BackgroundMutual influence between language and perception in multi-agent communication games
Ohmer, X., Marino, M., Franke, M., & König, P. (2022) · PLOS Computational Biology · doi:10.1371/journal.pcbi.1010658 · Cited by 6 · see citing works
“On a basic level, also the role of attention or memory could be studied with our setup, for example by using neural network agents with attention mechanisms [90] or external memory [91].”
Semi-supervised multimodal representation learning through a global workspace
Devillers, B., Maytié, L., & VanRullen, R. (2025) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2024.3416701 · Cited by 5 · see citing works
Episodic Memory in AI Agents Poses Risks that Should be Studied and Mitigated
DeChant, C. (2025) · 2025 IEEE Conference on Secure and Trustworthy Machine Learning (SaTML) · doi:10.1109/satml64287.2025.00024 · Cited by 5 · see citing works
“These included Long Short Term Memory modules [28], Neural Turing Machines [29], Differentiable Neural Computers [30], Hopfield networks [31], and Modern Hopfield Networks [32].”
Tomašev, N., Glorot, X., Rae, J. W., Zielinski, M., Askham, H., Saraiva, A., Mottram, A., Meyer, C., Ravuri, S., Protsyuk, I., et al. (2019) · doi:10.21203/rs.2.10083/v1 · Cited by 3 · see citing works
Self-citationBerges, V.-P., Oğuz, B., Haziza, D., Yih, W.-T., Zettlemoyer, L., & Ghosh, G. (2024) · arXiv · doi:10.48550/arxiv.2412.09764 · Cited by 1 · see citing works
“Neural Turing Machines (Graves et al., 2014, 2016) combine external trainable memory with other components to build a neural trainable computer.”
Zhou, S., & Li, Q. (2025) · Scientific Reports · doi:10.1038/s41598-025-31685-9
“These architectures implement content-based and location-based addressing mechanisms through differentiable read and write operations, formulated as: where M t denotes the memory matrix, w rt and w wt represent read and write attention weights, and a t is the add vector 29 .”
SEDM: Scalable Self-Evolving Distributed Memory for Agents
Xu, H., Hu, J., Zhang, K., Yu, L., Tang, Y., Song, X., Duan, Y., Ai, L., & Shi, B. (2025) · arXiv · doi:10.48550/arxiv.2509.09498
“Memory-augmented neural networks [13] and differentiable neural dictionaries [19] further demonstrate how structured memory access can enhance reasoning and generalization.”
Memory for Autonomous LLM Agents:Mechanisms, Evaluation, and Emerging Frontiers
Du, P. (2026) · arXiv · Cited by 1 · see citing works
“Neural Turing Machines [Graves et al., 2014] and Differentiable Neural Computers [Graves et al., 2016] pushed further, supporting both content-based and location-based addressing over a memory matrix.”
ARTEM: Enhancing Large Language Model Agents with Spatial-Temporal Episodic Memory
Tan, C. H.-M., Subagdja, B., & Tan, A.-H. (2026) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v40i30.39773
Kummara, H. (2026) · Cambridge Open Engage (preprint) · doi:10.33774/coe-2026-xpkvb
Memory Fabric for Conversational AI Agents: Enabling Shared and Persistent Memory Across Users
Tiwari, A., & Gupta, V. (2025) · TechRxiv · doi:10.36227/techrxiv.176523350.08289935/v1
On The Statistical Limits of Self-Improving Agents
Wang, C. L., Dorchen, K., & Jin, P. (2025) · arXiv · doi:10.48550/arxiv.2510.04399
- Authors
- Graves, A., Wayne, G., Reynolds, M., Harley, T., Danihelka, I., Grabska-Barwińska, A., Colmenarejo, S. G., Grefenstette, E., Ramalho, T., Agapiou, J., Badia, A. P., Hermann, K. M., Zwols, Y., Ostrovski, G., Cain, A., King, H., Summerfield, C., Blunsom, P., Kavukcuoglu, K., & Hassabis, D.
- Affiliations
- Google DeepMind (United Kingdom) (GB)
- Corresponding author
- Graves, Alex
- Wayne, Greg
- Venue
- Nature
- Published
- 2016-10-12
- Volume
- 538
- Issue
- 7626
- Pages
- 471-476
- ISSN
- 0028-0836 (ISSN-L)
- 1476-4687
- OpenAlex
- W2530887700
- Cited by
- 1,412
- References
- 8
- Citations by year
- 2026: 76 · 2025: 115 · 2024: 94 · 2023: 115 · 2022: 123 · 2021: 209 · 2020: 200 · 2019: 189 · 2018: 173 · 2017: 108 · 2016: 9
- FWCI
- 131.23
- Type
- journal article
- Language
- en
- Open access
- closed
- Licence
- Locations
- Nature (publishedVersion)
- PubMed (publishedVersion)
- RePEc: Research Papers in Economics (submittedVersion)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Neural Networks and Applications
- Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
- Physical Sciences › Computer Science › Artificial Intelligence › Neural Networks and Reservoir Computing
- Keywords
- external memory, random access memory, structured data representation, natural language reasoning, shortest path search, graph inference, reinforcement learning, supervised learning, transportation networks, family trees
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
