Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

A Path to Universal Neural Cellular Automata

T0 review · 4 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A single neural cellular automaton rule, trained by gradient descent on matrix primitives, can multiply and transpose matrices and emulate a neural network classifying MNIST directly in its state.

desk verdict A credible NCA training framework with a clean mutable/immutable hardware split and a clever MNIST block-decomposition demo, but the universal-computation narrative is ahead of the evidence. read the letter →

arxiv 2505.13058 v2 pith:EEXPC3ZQ submitted 2025-05-19 cs.LG cs.ETcs.NE

Gabriel Béna , Maxence Faldor , Dan F. M. Goodman , Antoine Cully This is my paper · ORCID
classification cs.LGcs.ETcs.NE
keywords NeuralCellularAutomataUniversalcomputationContinuousMatrixoperationsHardware-stateseparationMNISTemulationGradientdescentrulelearningEmergent
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Neural cellular automata—grids of cells updated by a learned local rule—are usually used for pattern formation. This paper argues they can also serve as a general computational substrate: one shared update rule, trained by gradient descent, is enough to perform matrix translation, multiplication, transposition, and rotation when paired with task-specific learned hardware patterns. The same trained rule can then emulate a pre-trained neural network solving MNIST by decomposing the matrix multiplication into small blocks carried out inside the automaton state, reaching about 60 percent accuracy against 84 percent for the original network. The result matters because a continuous cellular automaton that computes universally would connect analog computing with machine-learned rule discovery, moving beyond hand-designed discrete universal automata like the Game of Life.

What carries the argument

The load-bearing mechanism is the separation of the cell state into a mutable computational workspace and an immutable learned hardware state. The update rule is a neural network with a perception module (learnable convolution over the neighborhood) and an attention-based update module: a learned embedding of the hardware vector produces softmax attention weights over $N$ parallel MLP pathways, and the cell's state is updated residually by the attention-weighted mixture of those pathway outputs. The hardware can be monolithic, a full learned pattern over the grid, or modular, built from learned input, output, and task embedding vectors assembled per instance, and it is the hardware conditioning that lets the same rule switch between matrix operations. This two-level design, one general rule plus task-specific hardware, is what carries the empirical results.

What would settle it

Run a composite chain of, say, ten matrix multiplications, rotations, and translations executed sequentially with hardware reconfiguration between steps; if the final output error grows rapidly with chain length while each single operation is accurate, the composability claim fails.

Watch

Extended reading notes

Core claim

The central claim is that a single NCA update rule can learn a small algebra of matrix operations—identity mapping, multiplication, transposition, and rotation—across varied matrix sizes, placements, and input distributions, and that modular learned hardware lets this fixed rule generalize zero-shot to unseen task layouts. The strongest demonstration is the emulation of a linear MNIST classifier: matrix multiplication for a 784 × 10 weight matrix is split into 8 × 8 blocks, each computed in parallel in the cellular automaton state, then aggregated into logits. The emulated classifier reaches around 60 percent accuracy (the original reaches 84 percent) with about 69 percent prediction agreement, and the paper treats this accuracy loss as expected error propagation rather than a failure of the approach. On the paper's reading, these results are first steps toward a continuous universal cellular automaton and, eventually, analog general-purpose computation.

Load-bearing premise

The whole path to universal computation rests on the assumption that the learned update rule stays stable across sequential hardware reconfigurations, so that the output of one operation can feed the next without errors accumulating; the MNIST emulation, with its 84-to-60 percent accuracy drop, is the paper's own evidence that this stability is not yet there.

Editorial extensions

If this is right

  • If the central claim holds, one trained NCA rule can act as a shared, differentiable substrate for matrix algebra, so new tasks require learning only hardware patterns rather than retraining the dynamics.
  • Neural network inference can be embedded directly in the automaton's state, suggesting a route to physical or analog implementations where classification happens in the medium itself and no binary encoding is involved.
  • Modular hardware supports zero-shot generalization to unseen grid sizes, placements, and composite task chaining, so computational primitives can be assembled like components on a motherboard.
  • Error propagation is the main limiting factor: the MNIST emulation's drop from 84 to 60 percent accuracy is a direct measure of how much accuracy is lost when many block operations are aggregated.
  • These building blocks are the paper's stated foundation for a continuous universal cellular automaton, which would extend the universality of discrete automata to continuous dynamics.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper leaves implicit that the dual-timestep idea—fast cellular updates for computation, slow hardware reconfigurations for program flow—could develop into a compiler that maps algorithmic instructions onto learned hardware, but the sketches in Section 4.3 are not yet a demonstration.
  • A natural testable extension is to inject small state noise between chained operations to measure whether the learned rule has any error-correction property; if it does, composite tasks could scale well beyond the small illustrative examples shown.
  • The block-decomposition trick used for MNIST suggests that arbitrarily large matrix multiplications could be handled by a single fixed-size NCA rule, provided the per-block error is low enough; the 60 percent result sets a baseline for how much error currently accumulates.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper proposes a neural cellular automaton (NCA) architecture in which a mutable computational state is separated from an immutable, learned hardware state. A single attention-based update rule is trained by gradient descent on matrix primitives (translation/identity, multiplication, and rotation/transposition) using modular hardware embeddings for inputs, outputs, and task identity. The authors then show that a pretrained NCA multiplication rule, restricted to 8x8 blocks, can emulate a pretrained linear MNIST classifier through block-matrix decomposition, reaching about 60% accuracy versus 84% for the original classifier. Section 4.3 discusses chaining operations through hardware reconfiguration and outlines a planned graph-based hardware hypernetwork. The paper frames these results as a foundational step toward a continuous universal cellular automaton, not as a proof of universality.

Significance. If the reported results are reproducible, the main contribution is architectural: separating immutable hardware from mutable state and using modular hardware to obtain a single NCA rule that switches between computational modes. The MNIST emulation, although weak, is a nontrivial demonstration that a pretrained local rule can implement a global linear map via block decomposition. The paper is also candid about limitations: it acknowledges prior work by Whidden on NCA matrix multiplication, explicitly identifies the need for stability in chained tasks, and labels the hypernetwork as work in progress. What it does not provide is a quantitative demonstration of composability, which is the linchpin of the universality claim. No formal universality proof is claimed, and no code or full hyperparameters are provided.

major comments (4)
  1. [Section 4.2, Figure 5] The quantitative centerpiece is under-specified and contains an internal inconsistency: the text reports 84% original accuracy while the Figure 5 caption reports 86%, and no error bars, seeds, or statistical repetitions are given. More importantly, the 24-point accuracy drop is not decomposed: the paper never reports per-block 8x8 multiplication errors, nor how errors depend on block position, matrix distribution, or the aggregation step. Without this decomposition, the drop could be attributed to error accumulation across blocks, poor generalization of the multiplication primitive to MNIST-scale weight/pixel distributions, or artifacts of block aggregation, so the headline experiment is not diagnostic for the composability claim. I request seed variance and a per-stage error analysis.
  2. [Section 4.3, Figures 6 and 7] The path-to-universality argument relies on stable composition across sequential hardware reconfigurations, but this is never measured. The authors state that 'establishing stable representations becomes essential' and present Figure 7 as illustration, yet no success metrics, error rates, or end-to-end accuracy are reported for the chained tasks. The out-of-distribution task in Figure 6 also lacks quantitative evaluation. I request quantitative results for the chained sequences: per-stage output error and final-task success, together with a comparison against executing the same operations with ideal (non-NCA) primitives.
  3. [Section 3.5] Training details are insufficient for reproduction. The text does not report values for T_steps, the random final-step window, grid size, number of channels, number of attention pathways N, temperature T, optimizer, learning rate, batch composition, or training duration. The hardware fine-tuning speedup claim in Section 4.1.2 ('factor 2') is also stated without comparison conditions. Please add a hyperparameter table and the exact training protocol.
  4. [Section 4.1] The claim that the NCA 'mastered' matrix operations is supported only by qualitative statements and visual inspection; no numerical task errors are reported for translation, multiplication, or rotation under varied matrix sizes and distributions. Because the entire downstream argument depends on the reliability of these primitives, I ask for quantitative evaluation of each primitive, such as mean squared error or success rate as a function of matrix size and distribution.
minor comments (5)
  1. [Section 5] The conclusion refers to 'matrix multiplication and inversion', but the body trains transposition and rotation, not inversion; please correct the wording.
  2. [Figure 5 caption] The caption reports 'Original: 86% Emulated: 60%' while the text reports 84% for the original; please align the numbers.
  3. [Throughout] There are several typos, including 'implementions' in Section 3, 'developped' in Related Work, 'curently' in Section 4.3.1, and 'effetively' in the same section; a careful proofread is needed.
  4. [References] Some references are incomplete or informal, such as [30] (GitHub link without title or year) and [20] (web page without venue); please bring them into a consistent citation format.
  5. [Availability] No code or data availability statement is provided; releasing the training pipeline and evaluation scripts would substantially improve reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the MNIST emulation is a held-out application of independently trained components, and no central claim reduces to its own inputs.

full rationale

The paper's main quantitative result, the MNIST classifier emulation, is a genuine held-out application rather than a fitted prediction. The single-layer MLP is pre-trained on MNIST independently, and the NCA is pre-trained on 8x8 matrix multiplication tasks; the NCA is then applied without task-specific fine-tuning to block-decomposed MNIST matrix products, and the resulting logits are compared against the original classifier. No parameter in the NCA pipeline is fitted to the MNIST labels or to the target classification accuracy, so the 60%-versus-84% comparison is not forced by construction. The training of matrix primitives is direct supervised learning on those operations, not a prediction derived from a fitted input. The only self-citation is the CAX library [9] by two co-authors, which is used as an implementation tool and is not load-bearing for any theoretical claim. Section 4.3's composability discussion is explicitly framed as future directions and is supported only by illustrative figures, but lack of evidence for a speculative path is a correctness/evidence concern, not circularity. No equation defines a target in terms of an output, no fitted parameter is renamed as a prediction, and no uniqueness theorem is imported from the authors' prior work. The derivation chain is therefore self-contained as far as circularity is concerned.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

No new physical entities such as particles, forces, or conserved quantities are introduced. The immutable hardware is a learned embedding vector, not a new physical postulate, so the invented-entities list is empty.

free parameters (4)
  • Attention temperature T
    Softmax temperature controlling the sharpness of pathway selection in the update function; set by hand, value not reported.
  • Number of update steps (T_steps)
    Simulation length used in training and loss sampling; chosen by hand, exact value not given in the text.
  • Number of attention pathways (N)
    Number of parallel MLP pathways in the update function; architectural hyperparameter, not specified.
  • Grid size and channel count
    Spatial dimensions and hidden channel count of the NCA state; not reported, but the MNIST task relies on an 8 by 8 block size.
assumptions (4)
  • domain assumption Backpropagation through time can optimize the NCA to a stable solution for these tasks.
    The training relies on gradients through unrolled dynamics; standard in deep learning.
  • ad hoc to paper Attention gating over the hardware vector is sufficient to switch the same update rule between task-specific computational modes.
    Introduced in Section 3.2 without an ablation demonstrating this mechanism is necessary or superior.
  • standard math Decomposing the 784 by 10 matrix multiplication into 8 by 8 blocks and aggregating results reproduces the full product in exact arithmetic.
    Block matrix multiplication is exact in exact arithmetic; Section 4.2 applies this decomposition.
  • domain assumption The CAX library's implementation correctly computes the NCA dynamics.
    The paper depends on CAX without verifying its correctness in the text.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Path to Universal Neural Cellular Automata." pith.science (2026). https://pith.science/paper/EEXPC3ZQ

@misc{pith2026250513058,
  author       = {Pith},
  title        = {Pith review of: A Path to Universal Neural Cellular Automata},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EEXPC3ZQ}},
  note         = {Machine review of arXiv:2505.13058}
}
read the original abstract

Cellular automata have long been celebrated for their ability to generate complex behaviors from simple, local rules, with well-known discrete models like Conway's Game of Life proven capable of universal computation. Recent advancements have extended cellular automata into continuous domains, raising the question of whether these systems retain the capacity for universal computation. In parallel, neural cellular automata have emerged as a powerful paradigm where rules are learned via gradient descent rather than manually designed. This work explores the potential of neural cellular automata to develop a continuous Universal Cellular Automaton through training by gradient descent. We introduce a cellular automaton model, objective functions and training strategies to guide neural cellular automata toward universal computation in a continuous setting. Our experiments demonstrate the successful training of fundamental computational primitives - such as matrix multiplication and transposition - culminating in the emulation of a neural network solving the MNIST digit classification task directly within the cellular automata state. These results represent a foundational step toward realizing analog general-purpose computers, with implications for understanding universal computation in continuous dynamics and advancing the automated discovery of complex cellular automata behaviors via machine learning.

Figures

Figures reproduced from arXiv: 2505.13058 by the authors.

Figure 1
Figure 1. Our framework performs matrix operations (translation, multiplication, rotation) through continuous cellular [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Schematic of our architecture, showing the distinc [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Monolithic hardware configurations for 3 different [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Modular hardware configurations for 3 different [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: NCA emulates a neural network. Inputs show a batch of flattened MNIST images, alongside the weight matrix of [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Examples of an Out of Distribution Task: the NCA [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: Illustrating composite computational tasks using modular hardware configurations. Top panel: Multi-Rotation [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 8
Figure 8. Figure 8: Graph-based tasks representations and GNN-based [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

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

  1. Neural Field Turing Machine: A Differentiable Spatial Computer

    cs.NE 2025-08 reject novelty 4.0 of 10

    NFTM is a differentiable architecture with spatial read/write heads that claims Turing completeness and demonstrates learning local update rules for cellular automata, heat PDE, and CIFAR-10 inpainting.

Reference graph

Works this paper leans on

33 extracted references · 20 canonical work pages · cited by 1 Pith paper

  1. [1]

    Toward Artificial Open-Ended Evolution within Lenia using Quality-Diversity

    2024. Toward Artificial Open-Ended Evolution within Lenia using Quality-Diversity. Artificial Life Conference Proceedings, Vol. AL- IFE 2024: Proceedings of the 2024 Artificial Life Conference. doi:10.1162/isal_a_00827 arXiv:https://direct.mit.edu/isal/proceedings- pdf/isal2024/36/85/2461065/isal_a_00827.pdf

    doi:10.1162/isal_a_00827 2024

  2. [2]

    ADA University, Vladislav Gladkikh, Alexandr Nigay, and International Univer- sity of Information Technologies. 2018. Wireworld++: A Cellular Automaton for Simulation of Nonplanar Digital Electronic Circuits. 27, 1 (2018), 19–44. doi:10.25088/ComplexSystems.27.1.19

    work page doi:10.25088/complexsystems.27.1.19 2018

  3. [3]

    John Backus. 2007. Can programming be liberated from the von Neumann style? a functional style and its algebra of programs. Association for Computing Machinery, New York, NY, USA, 1977. https://doi.org/10.1145/1283920.1283933

    arXiv 2007

  4. [4]

    Danielle Smith Bassett and Ed Bullmore. 2006. Small-World Brain Net- works. The Neuroscientist 12, 6 (2006), 512–523. doi:10.1177/1073858406293182 arXiv:https://doi.org/10.1177/1073858406293182 PMID: 17079517

    work page doi:10.1177/1073858406293182 2006

  5. [5]

    Bullmore and Olaf Sporns

    Edward T. Bullmore and Olaf Sporns. 2012. The economy of brain network organization. Nature Reviews Neuroscience 13 (2012), 336–349. https://api. semanticscholar.org/CorpusID:16174225

    work page 2012

  6. [6]

    Bert Wang-Chak Chan. 2019. Lenia - Biology of Artificial Life. 28, 3 (2019), 251–286. doi:10.25088/ComplexSystems.28.3.251 arXiv:1812.05433 [nlin]

    arXiv 2019

  7. [7]

    Dennis Valbjørn Christensen, Regina Dittmann, Bernabé Linares-Barranco, Abu Sebastian, Manuel Le Gallo, Andrea Redaelli, Stefan Slesazeck, Thomas Mikolajick, Sabina Spiga, Stephan Menzel, I. I. Valov, Gianluca Milano, Carlo Ricciardi, Shi- Jun Liang, Feng Miao, Mario Lanza, Tyler James Quill, Scott Tom Keene, Alberto Salleo, Julie Grollier, Danijela Marko...

    work page 2021

  8. [8]

    Department of Computation and Neural Systems, Caltech, Mail Stop 136-93, Pasadena, California 91125, USA and Matthew Cook. 2004. Universality in Elementary Cellular Automata. 15, 1 (2004), 1–40. doi:10.25088/ComplexSystems. 15.1.1

    work page doi:10.25088/complexsystems 2004

Show all 33 references
  1. [9]

    Maxence Faldor and Antoine Cully. 2025. CAX: Cellular Automata Accelerated in JAX. In The Thirteenth International Conference on Learning Representations . https://openreview.net/forum?id=o2Igqm95SJ

    2025

  2. [10]

    Gary William Flake. 1998. The computational beauty of nature . MIT Press, Cambridge, MA, USA

    1998

  3. [11]

    Gajardo, A

    A. Gajardo, A. Moreira, and E. Goles. 2002. Complexity of Langton’s ant. 117, 1 (2002), 41–50. doi:10.1016/S0166-218X(00)00334-6

    2002 doi

  4. [12]

    Martin Gardner. 1970. MATHEMATICAL GAMES. Scientific American 223, 4 (1970), 120–123. http://www.jstor.org/stable/24927642

    1970

  5. [13]

    Etienne Guichard, Felix Reimers, Mia Kvalsund, Mikkel Lepperød, and Stefano Nichele. 2025. EngramNCA: a Neural Cellular Automaton Model of Memory Transfer. arXiv:2504.11855 [cs.NE] https://arxiv.org/abs/2504.11855

    2025 arXiv

  6. [14]

    Gautier Hamon, Mayalen Etcheverry, Bert Wang-Chak Chan, Clément Moulin- Frier, and Pierre-Yves Oudeyer. 2024. Discovering Sensorimotor Agency in Cellular Automata using Diversity Search. doi:10.48550/arXiv.2402.10236 arXiv:2402.10236 [cs]

    2024 doi arXiv

  7. [15]

    C. Mead. 1990. Neuromorphic electronic systems. Proc. IEEE 78, 10 (1990), 1629–1636. doi:10.1109/5.58356

    1990 doi

  8. [16]

    David Meunier, Renaud Lambiotte, and Edward Bullmore. 2010. Modular and Hierarchically Modular Organization of Brain Networks. 4 (12 2010), 200. doi:10. 3389/fnins.2010.00200

    2010 arXiv

  9. [17]

    Melanie Mitchell, James Crutchfield, and Rajarshi Das. 2000. Evolving Cellular Automata with Genetic Algorithms: A Review of Recent Work. First Int. Conf. on Evolutionary Computation and Its Applications 1 (May 2000)

    2000

  10. [18]

    Alexander Mordvintsev, Ettore Randazzo, Eyvind Niklasson, and Michael Levin

  11. [19]

    Elias Najarro, Shyam Sudhakaran, Claire Glanois, and Sebastian Risi. 2022. Hy- perNCA: Growing Developmental Networks with Neural Cellular Automata. doi:10.48550/arXiv.2204.11674 arXiv:2204.11674 [cs]

    2022 doi arXiv

  12. [20]

    Pietro Miotti, Eyvind Niklasson, Ettore Randazzo, Alexander Mordvintsev. 2025. Differentiable Logic CA: from Game of Life to Pattern Generation. https://google- research.github.io/self-organising-systems/difflogic-ca

    2025

  13. [21]

    Game of Life

    Stephan Rafler. 2011. Generalization of Conway’s "Game of Life" to a continuous domain - SmoothLife. doi:10.48550/arXiv.1111.1567 arXiv:1111.1567 [nlin]

    2011 doi arXiv

  14. [22]

    Ettore Randazzo, Alexander Mordvintsev, Eyvind Niklasson, Michael Levin, and Sam Greydanus. 2020. Self-classifying MNIST Digits. Distill (2020). doi:10.23915/ distill.00027.002 https://distill.pub/2020/selforg/mnist

    2020

  15. [23]

    Chris Reinke, Mayalen Etcheverry, and Pierre-Yves Oudeyer. 2020. Intrinsically Motivated Discovery of Diverse Patterns in Self-Organizing Systems. doi:10. 48550/arXiv.1908.06663 arXiv:1908.06663 [cs]

    2020 doi arXiv

  16. [24]

    Paul Rendell. 2016. Turing Machine Universality of the Game of Life . Emergence, Complexity and Computation, Vol. 18. Springer International Publishing. doi:10. 1007/978-3-319-19842-2

    2016

  17. [25]

    Emmanuel Sapin, Olivier Bailleux, and Jean-Jacques Chabrier. 2003. Research of a Cellular Automaton Simulating Logic Gates by Evolutionary Algorithms . doi:10. 1007/3-540-36599-0_39 Pages: 423

    2003

  18. [26]

    Schuman, Thomas E

    Catherine D. Schuman, Thomas E. Potok, Robert M. Patton, J. Douglas Birdwell, Mark E. Dean, Garrett S. Rose, and James S. Plank. 2017. A Survey of Neuromor- phic Computing and Neural Networks in Hardware. arXiv:1705.06963 [cs.NE] https://arxiv.org/abs/1705.06963

    2017 arXiv

  19. [27]

    A M Turing. 1937. ON COMPUTABLE NUMBERS, WITH AN APPLICATION TO THE ENTSCHEIDUNGSPROBLEM. (1937)

    1937

  20. [28]

    Bernd Ulmann. 2022. Analog Computing. De Gruyter Oldenbourg, Berlin, Boston. doi:doi:10.1515/9783110787740

    2022 doi

  21. [29]

    Weisstein

    Eric W. Weisstein. [n. d.]. Universal Cellular Automaton . https://mathworld. wolfram.com/UniversalCellularAutomaton.html Publisher: Wolfram Research, Inc

  22. [30]

    Peter Whidden. [n. d.]. . https://github.com/PWhiddy/Growing-Neural-Cellular- Automata-Pytorch/tree/master

  23. [31]

    Stephen Wolfram. 1984. Cellular automata as models of complexity. 311, 5985 (1984), 419–424. doi:10.1038/311419a0 Publisher: Nature Publishing Group

    1984 doi

  24. [32]

    Stephen Wolfram. 2002. A New Kind of Science . Wolfram Media. https://www. wolframscience.com 9

    2002

  25. [2020]

    Distill (2020)

    Growing Neural Cellular Automata. Distill (2020). doi:10.23915/distill.00023 https://distill.pub/2020/growing-ca

    2020 doi

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.