pith. machine review for the scientific record. sign in

arxiv: 1410.5401 · v2 · submitted 2014-10-20 · 💻 cs.NE

Recognition: 1 theorem link

· Lean Theorem

Neural Turing Machines

Alex Graves, Greg Wayne, Ivo Danihelka

Authors on Pith no claims yet

Pith reviewed 2026-05-13 07:29 UTC · model grok-4.3

classification 💻 cs.NE
keywords neural turing machinesexternal memoryattention mechanismsdifferentiable modelsalgorithm learningneural networksmemory augmented networks
0
0 comments X

The pith

Neural networks gain an external memory bank they control through soft attention, creating end-to-end differentiable systems that learn algorithms from examples.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper attaches a neural network controller to a large external memory matrix and lets the network read and write through differentiable attention mechanisms. Because every operation remains continuous, the whole architecture can be trained with gradient descent on input-output pairs alone. The resulting system learns to execute simple algorithmic tasks such as copying sequences, sorting numbers, and retrieving items by learned associations. This setup keeps the memory interactions smooth enough for back-propagation to adjust both the network weights and the attention patterns simultaneously.

Core claim

Neural Turing Machines combine a neural network controller with an external memory resource accessed by attentional read and write operations; the entire system is differentiable end-to-end and therefore trainable by gradient descent, allowing it to infer simple algorithms such as copying, sorting, and associative recall directly from example input-output pairs.

What carries the argument

Differentiable attentional read and write heads that interact with an external memory matrix.

Load-bearing premise

The soft attention operations used for reading and writing stay stable and trainable by gradient descent without causing vanishing gradients or optimization collapse on longer sequences.

What would settle it

Training runs that fail to converge on copying or sorting tasks once sequence length exceeds a modest threshold, with attention weights either collapsing or producing exploding gradients, would show the approach does not deliver stable algorithmic learning.

read the original abstract

We extend the capabilities of neural networks by coupling them to external memory resources, which they can interact with by attentional processes. The combined system is analogous to a Turing Machine or Von Neumann architecture but is differentiable end-to-end, allowing it to be efficiently trained with gradient descent. Preliminary results demonstrate that Neural Turing Machines can infer simple algorithms such as copying, sorting, and associative recall from input and output examples.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The paper introduces Neural Turing Machines (NTMs), neural networks augmented with an external differentiable memory matrix accessed via content-based and location-based attention heads. The controller is a neural network (feedforward or LSTM) that emits read/write weights; the full system is trained end-to-end by gradient descent. Preliminary experiments show the model can learn to copy, repeat-copy, sort, and perform associative recall on short synthetic sequences from input-output examples alone.

Significance. If the results hold under more rigorous evaluation, the work is significant because it supplies the first fully differentiable, end-to-end trainable analogue of a Turing machine with external memory. This opens a route to learning algorithmic procedures rather than merely pattern-matching, and the architecture has influenced subsequent memory-augmented networks. The paper also demonstrates that soft attention can implement both content and location addressing without hand-crafted rules.

major comments (3)
  1. [§4] §4 (Experiments) and associated figures: the abstract and text describe successful learning on copy, sort, and recall tasks but supply no numerical error rates, training curves, baseline comparisons (e.g., LSTM or RNN), or hyper-parameter details. Without these, the claim that NTMs “infer simple algorithms” cannot be quantitatively evaluated.
  2. [§3.2–3.3] §3.2–3.3 (Addressing mechanisms): the content and location addressing weights are produced by softmax; no analysis or ablation is given for the product of successive softmax Jacobians over many timesteps. This directly bears on the skeptic’s concern that gradients may vanish for sequences longer than the training lengths shown (~20 tokens).
  3. [§4.1] §4.1 (Copy and repeat-copy tasks): success is reported only on short fixed-length sequences; no test of generalization to lengths substantially beyond the training distribution is presented, which is load-bearing for the claim that the model learns a general copying algorithm rather than a finite-state pattern.
minor comments (2)
  1. [§3] Notation for the memory matrix M_t and the read vector r_t is introduced without an explicit equation number in the first occurrence; adding an equation label would improve readability.
  2. [§2] The paper cites only a handful of prior memory-augmented networks; a short related-work paragraph situating the NTM against contemporaneous differentiable-memory proposals would help readers.

Simulated Author's Rebuttal

3 responses · 0 unresolved

Thank you for your thoughtful review of our paper on Neural Turing Machines. We have carefully considered each of your major comments and have made revisions to the manuscript to address them where possible. Our point-by-point responses are provided below.

read point-by-point responses

  1. Referee: [§4] §4 (Experiments) and associated figures: the abstract and text describe successful learning on copy, sort, and recall tasks but supply no numerical error rates, training curves, baseline comparisons (e.g., LSTM or RNN), or hyper-parameter details. Without these, the claim that NTMs “infer simple algorithms” cannot be quantitatively evaluated.

    Authors: We agree that additional quantitative details would strengthen the presentation. In the revised manuscript we have added training curves for each task (showing convergence to near-zero error), reported explicit final error rates in the text, included LSTM and RNN baseline comparisons demonstrating superior performance by the NTM on algorithmic tasks, and moved all hyper-parameter settings to a new appendix. revision: yes

  2. Referee: [§3.2–3.3] §3.2–3.3 (Addressing mechanisms): the content and location addressing weights are produced by softmax; no analysis or ablation is given for the product of successive softmax Jacobians over many timesteps. This directly bears on the skeptic’s concern that gradients may vanish for sequences longer than the training lengths shown (~20 tokens).

    Authors: We acknowledge the value of analyzing gradient propagation through the successive softmax operations. Our empirical results show stable training without apparent vanishing for the lengths used; the combination of content-based and location-based addressing (with its convolutional shift) empirically preserves gradient flow. In revision we have added a short discussion in §3.3 on this point and the role of the shift operation, though a full Jacobian ablation remains future work. revision: partial

  3. Referee: [§4.1] §4.1 (Copy and repeat-copy tasks): success is reported only on short fixed-length sequences; no test of generalization to lengths substantially beyond the training distribution is presented, which is load-bearing for the claim that the model learns a general copying algorithm rather than a finite-state pattern.

    Authors: The original experiments already included tests on sequences longer than the training distribution to support the algorithmic claim. To make this explicit we have expanded §4.1 with new results on variable-length inputs up to twice the training length, confirming that error rates remain low and the model continues to execute the copying procedure correctly. revision: yes

Circularity Check

0 steps flagged

No significant circularity in architectural proposal and empirical validation

full rationale

The paper defines a novel Neural Turing Machine architecture by specifying controller, memory, and differentiable attentional read/write mechanisms, then validates it through experiments on synthetic tasks such as copying and associative recall. No derivation step reduces a claimed result to a fitted parameter or self-referential definition by construction. No load-bearing self-citations are used to establish uniqueness theorems or to smuggle in ansatzes. The central claims rest on explicit model equations and reported training outcomes rather than any circular reduction, making the work self-contained as an empirical architecture proposal.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 1 invented entities

The central claim rests on the assumption that memory read/write operations can be made fully differentiable and that gradient descent will successfully train the controller and attention mechanisms on the target tasks.

free parameters (1)
  • memory size and number of heads
    Architectural hyperparameters that determine the external memory dimensions and attention capacity; chosen per task.
axioms (1)
  • domain assumption All memory access operations are differentiable
    Required for end-to-end gradient descent but not proven in the abstract.
invented entities (1)
  • external memory bank with attention-based access no independent evidence
    purpose: To provide storage beyond the neural controller's internal state
    New component introduced by the paper; no independent evidence outside the reported experiments.

pith-pipeline@v0.9.0 · 5342 in / 1184 out tokens · 69239 ms · 2026-05-13T07:29:37.093817+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 30 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Gradient-Based Program Synthesis with Neurally Interpreted Languages

    cs.LG 2026-04 unverdicted novelty 8.0

    NLI autonomously discovers a vocabulary of primitive operations and interprets variable-length programs via a neural executor, allowing end-to-end training and gradient-based test-time adaptation that outperforms prio...

  2. On the Mirage of Long-Range Dependency, with an Application to Integer Multiplication

    cs.LG 2026-03 unverdicted novelty 8.0

    Long-range dependency in integer multiplication is a mirage from 1D representation; a 2D grid reduces it to local 3x3 operations, letting a 321-parameter neural cellular automaton generalize perfectly to inputs 683 ti...

  3. RULER: What's the Real Context Size of Your Long-Context Language Models?

    cs.CL 2024-04 accept novelty 8.0

    RULER shows most long-context LMs drop sharply in performance on complex tasks as length and difficulty increase, with only half maintaining results at 32K tokens.

  4. Grokking: Generalization Beyond Overfitting on Small Algorithmic Datasets

    cs.LG 2022-01 unverdicted novelty 8.0

    Neural networks exhibit grokking on small algorithmic datasets, achieving perfect generalization well after overfitting.

  5. Show Your Work: Scratchpads for Intermediate Computation with Language Models

    cs.LG 2021-11 unverdicted novelty 8.0

    Training language models to generate intermediate computation steps on a scratchpad enables them to perform multi-step tasks such as long addition and arbitrary program execution that they otherwise fail at.

  6. Categorical Reparameterization with Gumbel-Softmax

    stat.ML 2016-11 unverdicted novelty 8.0

    Gumbel-Softmax provides a continuous relaxation of categorical sampling that anneals to discrete samples for gradient-based optimization.

  7. Adaptive Computation Time for Recurrent Neural Networks

    cs.NE 2016-03 accept novelty 8.0

    ACT lets RNNs dynamically adapt computation depth per input via a differentiable halting unit, yielding large gains on synthetic tasks and structural insights on language data.

  8. Does Engram Do Memory Retrieval in Autoregressive Image Generation?

    cs.CV 2026-05 accept novelty 7.0

    Engram in AR image generation saves backbone FLOPs but trails pure AR baselines in FID and behaves as a gated side-pathway rather than a content-addressed retriever.

  9. Intrinsic Vicarious Conditioning for Deep Reinforcement Learning

    cs.LG 2026-05 unverdicted novelty 7.0

    Vicarious conditioning is proposed as a new intrinsic reward in RL that implements attention, retention, reproduction, and reinforcement via memory methods to enable low-shot learning from others without their policie...

  10. On the Importance of Multistability for Horizon Generalization in Reinforcement Learning

    cs.LG 2026-05 unverdicted novelty 7.0

    Multistability is necessary for temporal horizon generalization in POMDPs, sufficient in simple tasks along with transient dynamics in complex ones, while monostable parallelizable RNNs like SSMs and gated linear RNNs...

  11. Neural Information Causality

    quant-ph 2026-05 unverdicted novelty 7.0

    Neural-IC separates embedding inequalities from capacity bounds in query-separated computations, with one-bit RAC benchmarks and CHSH-layer stability selecting the Tsirelson threshold for quantum enhancements.

  12. Sharp Capacity Thresholds in Linear Associative Memory: From Winner-Take-All to Listwise Retrieval

    stat.ML 2026-05 unverdicted novelty 7.0

    Winner-take-all linear memory capacity scales as d² ~ n log n due to extreme values; listwise retrieval via Tail-Average Margin yields d² ~ n with exact asymptotic theory.

  13. Screening Is Enough

    cs.LG 2026-04 unverdicted novelty 7.0

    Multiscreen replaces softmax attention with screening to provide absolute query-key relevance, resulting in models with 30% fewer parameters that maintain stable performance at long contexts.

  14. Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach

    cs.LG 2025-02 unverdicted novelty 7.0

    A recurrent-depth architecture enables language models to improve reasoning performance by iterating computation in latent space, achieving gains equivalent to much larger models on benchmarks.

  15. Concrete Problems in AI Safety

    cs.AI 2016-06 accept novelty 7.0

    The paper categorizes five concrete AI safety problems arising from flawed objectives, costly evaluation, and learning dynamics.

  16. Phasor Memory Networks: Stable Backpropagation Through Time for Scalable Explicit Memory

    cs.LG 2026-05 unverdicted novelty 6.0

    PMNet uses unitary phasor dynamics and hierarchical anchors to make explicit memory stable for long sequences, matching a 3x larger Mamba model on long-context robustness with a 119M parameter network.

  17. The Position Curse: LLMs Struggle to Locate the Last Few Items in a List

    cs.LG 2026-05 unverdicted novelty 6.0

    LLMs exhibit the Position Curse, with backward position retrieval in lists lagging far behind forward retrieval, showing only partial gains from PosBench fine-tuning.

  18. FAAST: Forward-Only Associative Learning via Closed-Form Fast Weights for Test-Time Supervised Adaptation

    cs.LG 2026-05 unverdicted novelty 6.0

    FAAST analytically compiles labeled examples into fast weights via a single forward pass, matching backprop adaptation performance with over 90% less time and up to 95% less memory than memory-based methods.

  19. Borrowed Geometry: Computational Reuse of Frozen Text-Pretrained Transformer Weights Across Modalities

    cs.LG 2026-05 unverdicted novelty 6.0

    Frozen text-pretrained transformer weights transfer across modalities through a thin interface, achieving SOTA on a robotic task and parity on decision-making with far fewer trainable parameters.

  20. Universal Transformers Need Memory: Depth-State Trade-offs in Adaptive Recursive Reasoning

    cs.LG 2026-04 conditional novelty 6.0

    Memory tokens are required for non-trivial performance in adaptive Universal Transformers on Sudoku-Extreme, with 8-32 tokens yielding stable 57% exact-match accuracy while trading off against ponder depth.

  21. Ask Only When Needed: Proactive Retrieval from Memory and Skills for Experience-Driven Lifelong Agents

    cs.CL 2026-04 unverdicted novelty 6.0

    ProactAgent learns a proactive retrieval policy via reinforcement learning on paired task continuations, improving lifelong agent performance and cutting retrieval overhead on SciWorld, AlfWorld, and StuLife.

  22. Titans: Learning to Memorize at Test Time

    cs.LG 2024-12 unverdicted novelty 6.0

    Titans combine attention for current context with a learnable neural memory for long-term history, achieving better performance and scaling to over 2M-token contexts on language, reasoning, genomics, and time-series tasks.

  23. Geometric Deep Learning: Grids, Groups, Graphs, Geodesics, and Gauges

    cs.LG 2021-04 accept novelty 6.0

    Geometric deep learning provides a unified mathematical framework based on grids, groups, graphs, geodesics, and gauges to explain and extend neural network architectures by incorporating physical regularities.

  24. Universal Transformers

    cs.CL 2018-07 unverdicted novelty 6.0

    Universal Transformers combine Transformer parallelism with recurrent updates and dynamic halting to achieve Turing-completeness under assumptions and outperform standard Transformers on algorithmic and language tasks.

  25. TIDE: Every Layer Knows the Token Beneath the Context

    cs.CL 2026-05 unverdicted novelty 5.0

    TIDE augments standard transformers with per-layer token embedding injection via an ensemble of memory blocks and a depth-conditioned router to mitigate rare-token undertraining and contextual collapse.

  26. FAAST: Forward-Only Associative Learning via Closed-Form Fast Weights for Test-Time Supervised Adaptation

    cs.LG 2026-05 unverdicted novelty 5.0

    FAAST performs test-time supervised adaptation by analytically deriving fast weights from examples in one forward pass, matching backprop performance with over 90% less adaptation time and up to 95% memory savings ver...

  27. Graph Memory Transformer (GMT)

    cs.LG 2026-04 unverdicted novelty 5.0

    Graph Memory Transformer (GMT) swaps dense FFN sublayers for a graph of 128 centroids and a learned 128x128 transition matrix per block, yielding a 82M-parameter decoder-only LM that trains stably but trails a 103M de...

  28. Neural Computers

    cs.LG 2026-04 unverdicted novelty 5.0

    Neural Computers are introduced as a new machine form where computation, memory, and I/O are unified in a learned runtime state, with initial video-model experiments showing acquisition of basic interface primitives f...

  29. Event-Centric World Modeling with Memory-Augmented Retrieval for Embodied Decision-Making

    cs.LG 2026-04 unverdicted novelty 4.0

    An event-centric framework encodes environments as semantic events and retrieves weighted prior maneuvers from a knowledge bank to enable interpretable, physics-aware decision-making for UAVs.

  30. A PyTorch Library of Turing-Complete Neural Networks

    cs.LG 2026-05 unverdicted novelty 3.0

    A PyTorch package constructs neural networks that exactly simulate given Turing machines using transformer and recurrent architectures derived from prior theoretical results.

Reference graph

Works this paper leans on

42 extracted references · 42 canonical work pages · cited by 29 Pith papers · 1 internal anchor

  1. [1]

    Baddeley, A., Eysenck, M., and Anderson, M. (2009). Memory . Psychology Press

    work page 2009

  2. [2]

    Bahdanau, D., Cho, K., and Bengio, Y. (2014). Neural machine translation by jointly learning to align and translate. abs/1409.0473

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  3. [3]

    Barrouillet, P., Bernardin, S., and Camos, V. (2004). Time constraints and resource sharing in adults' working memory spans. Journal of Experimental Psychology: General , 133(1):83

    work page 2004

  4. [4]

    Chomsky, N. (1956). Three models for the description of language. Information Theory, IEEE Transactions on , 2(3):113--124

    work page 1956

  5. [5]

    L., and Sun, G.-Z

    Das, S., Giles, C. L., and Sun, G.-Z. (1992). Learning context-free grammars: Capabilities and limitations of a recurrent neural network with an external stack memory. In Proceedings of The Fourteenth Annual Conference of Cognitive Science Society. Indiana University

    work page 1992

  6. [6]

    Dayan, P. (2008). Simple substrates for complex cognition. Frontiers in neuroscience , 2(2):255

    work page 2008

  7. [7]

    Eliasmith, C. (2013). How to build a brain: A neural architecture for biological cognition . Oxford University Press

    work page 2013

  8. [8]

    D., and Chomsky, N

    Fitch, W., Hauser, M. D., and Chomsky, N. (2005). The evolution of the language faculty: clarifications and implications. Cognition , 97(2):179--210

    work page 2005

  9. [9]

    Fodor, J. A. and Pylyshyn, Z. W. (1988). Connectionism and cognitive architecture: A critical analysis. Cognition , 28(1):3--71

    work page 1988

  10. [10]

    Frasconi, P., Gori, M., and Sperduti, A. (1998). A general framework for adaptive processing of data structures. Neural Networks, IEEE Transactions on , 9(5):768--786

    work page 1998

  11. [11]

    Gallistel, C. R. and King, A. P. (2009). Memory and the computational brain: Why cognitive science will transform neuroscience , volume 3. John Wiley & Sons

    work page 2009

  12. [12]

    Goldman-Rakic, P. S. (1995). Cellular basis of working memory. Neuron , 14(3):477--485

    work page 1995

  13. [13]

    Graves, A. (2013). Generating sequences with recurrent neural networks. arXiv preprint arXiv:1308.0850

    work page arXiv 2013

  14. [14]

    and Jaitly, N

    Graves, A. and Jaitly, N. (2014). Towards end-to-end speech recognition with recurrent neural networks. In Proceedings of the 31st International Conference on Machine Learning (ICML-14) , pages 1764--1772

    work page 2014

  15. [15]

    Graves, A., Mohamed, A., and Hinton, G. (2013). Speech recognition with deep recurrent neural networks. In Acoustics, Speech and Signal Processing (ICASSP), 2013 IEEE International Conference on , pages 6645--6649. IEEE

    work page 2013

  16. [16]

    Hadley, R. F. (2009). The problem of rapid variable creation. Neural computation , 21(2):510--532

    work page 2009

  17. [17]

    E., Frank, M

    Hazy, T. E., Frank, M. J., and O'Reilly, R. C. (2006). Banishing the homunculus: making working memory work. Neuroscience , 139(1):105--118

    work page 2006

  18. [18]

    Hinton, G. E. (1986). Learning distributed representations of concepts. In Proceedings of the eighth annual conference of the cognitive science society , volume 1, page 12. Amherst, MA

    work page 1986

  19. [19]

    Hochreiter, S., Bengio, Y., Frasconi, P., and Schmidhuber, J. (2001a). Gradient flow in recurrent nets: the difficulty of learning long-term dependencies

    work page

  20. [20]

    and Schmidhuber, J

    Hochreiter, S. and Schmidhuber, J. (1997). Long short-term memory. Neural computation , 9(8):1735--1780

    work page 1997

  21. [21]

    S., and Conwell, P

    Hochreiter, S., Younger, A. S., and Conwell, P. R. (2001b). Learning to learn using gradient descent. In Artificial Neural Networks?ICANN 2001 , pages 87--94. Springer

    work page 2001

  22. [22]

    Hopfield, J. J. (1982). Neural networks and physical systems with emergent collective computational abilities. Proceedings of the national academy of sciences , 79(8):2554--2558

    work page 1982

  23. [23]

    and Pinker, S

    Jackendoff, R. and Pinker, S. (2005). The nature of the language faculty and its implications for evolution of language (reply to fitch, hauser, and chomsky). Cognition , 97(2):211--225

    work page 2005

  24. [24]

    Kanerva, P. (2009). Hyperdimensional computing: An introduction to computing in distributed representation with high-dimensional random vectors. Cognitive Computation , 1(2):139--159

    work page 2009

  25. [25]

    Marcus, G. F. (2003). The algebraic mind: Integrating connectionism and cognitive science . MIT press

    work page 2003

  26. [26]

    Miller, G. A. (1956). The magical number seven, plus or minus two: some limits on our capacity for processing information. Psychological review , 63(2):81

    work page 1956

  27. [27]

    Miller, G. A. (2003). The cognitive revolution: a historical perspective. Trends in cognitive sciences , 7(3):141--144

    work page 2003

  28. [28]

    Minsky, M. L. (1967). Computation: finite and infinite machines . Prentice-Hall, Inc

    work page 1967

  29. [29]

    Murphy, K. P. (2012). Machine learning: a probabilistic perspective . MIT press

    work page 2012

  30. [30]

    Plate, T. A. (2003). Holographic Reduced Representation: Distributed representation for cognitive structures . CSLI

    work page 2003

  31. [31]

    Pollack, J. B. (1990). Recursive distributed representations. Artificial Intelligence , 46(1):77--105

    work page 1990

  32. [32]

    R., Wang, X.-J., Daw, N

    Rigotti, M., Barak, O., Warden, M. R., Wang, X.-J., Daw, N. D., Miller, E. K., and Fusi, S. (2013). The importance of mixed selectivity in complex cognitive tasks. Nature , 497(7451):585--590

    work page 2013

  33. [33]

    E., McClelland, J

    Rumelhart, D. E., McClelland, J. L., Group, P. R., et al. (1986). Parallel distributed processing , volume 1. MIT press

    work page 1986

  34. [34]

    Seung, H. S. (1998). Continuous attractors and oculomotor control. Neural Networks , 11(7):1253--1258

    work page 1998

  35. [35]

    Siegelmann, H. T. and Sontag, E. D. (1995). On the computational power of neural nets. Journal of computer and system sciences , 50(1):132--150

    work page 1995

  36. [36]

    Smolensky, P. (1990). Tensor product variable binding and the representation of symbolic structures in connectionist systems. Artificial intelligence , 46(1):159--216

    work page 1990

  37. [37]

    D., and Ng, A

    Socher, R., Huval, B., Manning, C. D., and Ng, A. Y. (2012). Semantic compositionality through recursive matrix-vector spaces. In Proceedings of the 2012 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning , pages 1201--1211. Association for Computational Linguistics

    work page 2012

  38. [38]

    Sutskever, I., Martens, J., and Hinton, G. E. (2011). Generating text with recurrent neural networks. In Proceedings of the 28th International Conference on Machine Learning (ICML-11) , pages 1017--1024

    work page 2011

  39. [39]

    Sutskever, I., Vinyals, O., and Le, Q. V. (2014). Sequence to sequence learning with neural networks. arXiv preprint arXiv:1409.3215

    work page arXiv 2014

  40. [40]

    Touretzky, D. S. (1990). Boltzcons: Dynamic symbol structures in a connectionist network. Artificial Intelligence , 46(1):5--46

    work page 1990

  41. [41]

    Von Neumann, J. (1945). First draft of a report on the edvac

    work page 1945

  42. [42]

    Wang, X.-J. (1999). Synaptic basis of cortical persistent activity: the importance of nmda receptors to working memory. The Journal of Neuroscience , 19(21):9587--9603

    work page 1999