pith. sign in

arxiv: 2512.17129 · v2 · submitted 2025-12-18 · 💻 cs.LG · cs.MA· cs.RO· q-bio.QM

DiffeoMorph: Learning to Morph 3D Shapes Using Differentiable Agent-Based Simulations

Seong Ho Pahng , Guoye Guan , Benjamin Fefferman , Sahand Hormoz This is my paper

Pith reviewed 2026-05-16 21:01 UTC · model grok-4.3

classification 💻 cs.LG cs.MAcs.ROq-bio.QM
keywords DiffeoMorphagent-based morphogenesisdifferentiable simulationZernike polynomialsSE(3)-equivariant GNNdistributed control3D shape formationswarm robotics
0
0 comments X

The pith

DiffeoMorph trains shared agent rules so populations self-organize into complex 3D target shapes from minimal initial patterns.

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

The paper introduces an end-to-end differentiable framework in which each agent follows the same update rule based on an SE(3)-equivariant graph neural network that processes its internal state and signals from neighbors. Training relies on a shape-matching loss that represents both predicted and target forms as continuous distributions via 3D Zernike polynomials, making the loss invariant to agent ordering, count, and global orientation after an optimal alignment rotation. This setup lets the system learn distributed control protocols that achieve precise global morphologies without central coordination. A sympathetic reader would care because the same mechanism could illuminate how biological collectives produce reliable structures and could supply design rules for robotic swarms or programmable materials.

Core claim

DiffeoMorph is an end-to-end differentiable framework for learning a morphogenesis protocol in which agents update position and internal state through an SE(3)-equivariant graph neural network; the system is trained by a new loss that compares predicted and target shapes as continuous spatial distributions expressed with 3D Zernike polynomials, after an alignment step that rotates the predicted spectrum to best match the target while preserving reflection sensitivity, thereby enabling formation of a range of complex 3D shapes from minimally patterned initial conditions.

What carries the argument

SE(3)-equivariant graph neural network for local agent updates together with the 3D Zernike polynomial loss plus optimal rotation alignment that compares shapes as continuous distributions.

If this is right

  • Distributed control strategies can be learned that achieve precise global 3D patterns from local interactions alone.
  • Shape comparison becomes independent of agent ordering, number, and orientation, allowing flexible population sizes.
  • The same framework supplies a general method for training swarm behaviors in robotics and self-assembly systems.
  • Benchmarking shows the Zernike loss outperforms standard point-cloud distance metrics on shape-matching tasks.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the simulation faithfully captures physical interactions, the learned protocols could be transferred to real-world robotic or biological experiments.
  • The approach could be extended to growing or dividing agent populations by making the graph neural network handle variable node counts during training.
  • Connections emerge between multi-agent learning and classical models of pattern formation that rely on continuous reaction-diffusion fields.

Load-bearing premise

The learned protocols produced by the differentiable Zernike-loss simulation will continue to work when applied to shapes or agent counts outside the specific training demonstrations.

What would settle it

Train a protocol on one collection of target shapes and a fixed agent count, then test whether the identical protocol produces the correct morphology for an untrained target shape or with a substantially different number of agents.

Figures

Figures reproduced from arXiv: 2512.17129 by Benjamin Fefferman, Guoye Guan, Sahand Hormoz, Seong Ho Pahng.

Figure 1
Figure 1. Figure 1: FIG. 1. Overview of the morphogenesis model and shape optimization of DiffeoMorph. ( [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2. Spectral alignment and shape optimization by matching 3D Zernike moments. ( [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3. Benchmarking the proposed loss. ( [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4. Visualization of morphogenesis trajectories from trained models. ( [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5. Visualization of the evolution of internal states during morphogenesis. ( [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗
read the original abstract

Biological systems can form complex three-dimensional structures through the collective behavior of agents that share a common update rule and operate without central control. How such distributed control gives rise to precise global patterns remains a central question not only in developmental biology but also in distributed robotics, programmable matter, and multi-agent learning. Here, we introduce DiffeoMorph, an end-to-end differentiable framework for learning a morphogenesis protocol that guides a population of agents to morph into a target 3D shape. Each agent updates its position and internal state using an SE(3)-equivariant graph neural network, based on its own internal state and signals received from other agents. To train this system, we introduce a new shape-matching loss based on 3D Zernike polynomials, which compares the predicted and target shapes as continuous spatial distributions, not as discrete point clouds, and is invariant to agent ordering, number of agents, and global orientation. To achieve rotation invariance while preserving reflection sensitivity, we include an alignment step that optimally rotates the predicted Zernike spectrum to match the target before computing the loss. We perform benchmarking to establish the advantages of our shape-matching loss over other standard distance metrics for shape comparison tasks. We then demonstrate that DiffeoMorph can form a range of complex shapes from minimally patterned initial conditions. DiffeoMorph provides a general framework for learning distributed control strategies for morphogenesis, swarm robotics, and programmable self-assembly.

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

2 major / 2 minor

Summary. The paper introduces DiffeoMorph, an end-to-end differentiable framework in which agents update positions and states via an SE(3)-equivariant GNN to form target 3D shapes from minimal initial conditions. Training uses a novel Zernike-polynomial shape-matching loss that treats shapes as continuous distributions and is invariant to agent count, ordering, and global orientation (with an explicit alignment step for rotation invariance). The manuscript reports benchmarking of this loss against standard metrics and demonstrations that the system can produce a range of complex morphologies.

Significance. If the central claims hold, the work provides a general, differentiable approach to learning distributed control protocols for morphogenesis and self-assembly. The Zernike loss and SE(3)-equivariant architecture are concrete technical contributions that address invariance issues in multi-agent shape formation; successful transfer of learned protocols would have clear implications for swarm robotics and programmable matter.

major comments (2)
  1. [Experiments] Experiments section: the reported demonstrations are confined to the same shapes used during training; no held-out shape tests, cross-shape protocol transfer metrics, or agent-count sweeps are described. This leaves the central claim—that the GNN discovers transferable update rules rather than per-shape policies—unsupported by the presented evidence.
  2. [§4] §4 (demonstrations): without quantitative metrics on generalization (e.g., success rate on unseen target shapes or different N), it remains possible that the learned policies overfit the training morphologies, undermining the assertion of a general framework for distributed morphogenesis.
minor comments (2)
  1. [Abstract] The abstract states that benchmarking establishes advantages of the Zernike loss but does not name the competing metrics or report quantitative differences; this detail should be added for clarity.
  2. [Method] Notation for the Zernike spectrum alignment step is introduced without an explicit equation reference; adding a numbered equation would improve reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. The comments correctly identify that the current experimental section would benefit from explicit generalization tests. We address each major comment below and will revise the manuscript to incorporate the suggested experiments.

read point-by-point responses

  1. Referee: [Experiments] Experiments section: the reported demonstrations are confined to the same shapes used during training; no held-out shape tests, cross-shape protocol transfer metrics, or agent-count sweeps are described. This leaves the central claim—that the GNN discovers transferable update rules rather than per-shape policies—unsupported by the presented evidence.

    Authors: We agree that the presented demonstrations use shapes from the training distribution and that held-out evaluations are needed to substantiate transferability. In the revised manuscript we will add a dedicated generalization section that includes: (i) training on a subset of target morphologies and reporting success rates on held-out shapes, (ii) cross-shape protocol transfer metrics, and (iii) agent-count sweeps across different values of N. These additions will directly support the claim that the SE(3)-equivariant GNN learns distributed update rules that generalize rather than memorizing per-shape policies. revision: yes

  2. Referee: [§4] §4 (demonstrations): without quantitative metrics on generalization (e.g., success rate on unseen target shapes or different N), it remains possible that the learned policies overfit the training morphologies, undermining the assertion of a general framework for distributed morphogenesis.

    Authors: We acknowledge that the current version lacks quantitative generalization metrics, which leaves open the possibility of overfitting. The revision will include success-rate tables for unseen target shapes and for varying agent counts N, together with the corresponding protocol-transfer experiments. These quantitative results will strengthen the evidence that DiffeoMorph provides a general framework rather than shape-specific solutions. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper defines an independent SE(3)-equivariant GNN update rule and a new Zernike-polynomial shape-matching loss that operates on continuous distributions, invariant to agent count and ordering. Training is end-to-end differentiable against target shapes, with explicit benchmarking of the loss against standard metrics. No equations reduce the target morphology or learned protocols to quantities fitted from the same data; no self-citations are load-bearing for the central claims; the architecture and loss are introduced as novel components rather than derived from prior author work by construction. The framework remains self-contained against external shape-comparison benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 1 invented entities

The central claim rests on the learned GNN parameters and the effectiveness of the Zernike loss; the framework assumes SE(3) equivariance holds for physical agent interactions and that the loss provides a sufficient training signal.

free parameters (1)
  • GNN weights and biases
    Learned via gradient descent on the shape-matching loss; these define the update rule for each agent.
axioms (1)
  • domain assumption SE(3)-equivariance of the graph neural network update rule
    Invoked to ensure agent updates respect 3D rotations and translations without additional training data.
invented entities (1)
  • Zernike-polynomial shape representation for loss no independent evidence
    purpose: To compare predicted and target shapes as continuous distributions invariant to ordering and orientation
    New application of Zernike polynomials as a differentiable loss in this agent-based setting; no independent evidence provided beyond the described benchmarking.

pith-pipeline@v0.9.0 · 5579 in / 1305 out tokens · 176762 ms · 2026-05-16T21:01:42.459366+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

79 extracted references · 79 canonical work pages · 2 internal anchors

  1. [1]

    organizer cells

    to compute spherical harmonicsY ℓm. We usevmapfunction to vectorize the projection over eachn,ℓ, andm in equation (5). Since the shape of tensors passed tovmapshould be identical, for eachℓ-row ofY ℓm, we assign the expression of them-th real spherical harmonics starting from the first column and zero-padded the azimuthalm indices after 2ℓ+1 up to 2ℓ max+...

    work page 2000

  2. [2]

    S. F. Gilbert and M. J. F. Barresi,Developmental Biology, 12th ed. (Sinauer Associates, Oxford University Press, New York, NY, 2020)

    work page 2020

  3. [3]

    Hogeweg, Evolving mechanisms of morphogenesis: on the interplay between differential adhesion and cell differentiation, Journal of Theoretical Biology203, 317 (2000)

    P. Hogeweg, Evolving mechanisms of morphogenesis: on the interplay between differential adhesion and cell differentiation, Journal of Theoretical Biology203, 317 (2000)

    work page 2000

  4. [4]

    B. P. Teague, P. Guye, and R. Weiss, Synthetic morphogenesis, Cold Spring Harbor perspectives in biology8, a023929 (2016)

    work page 2016

  5. [5]

    Takeichi, Dynamic contacts: rearranging adherens junctions to drive epithelial remodelling, Nature reviews Molecular cell biology15, 397 (2014)

    M. Takeichi, Dynamic contacts: rearranging adherens junctions to drive epithelial remodelling, Nature reviews Molecular cell biology15, 397 (2014)

    work page 2014

  6. [6]

    J. B. Gurdon and P.-Y. Bourillot, Morphogen gradient interpretation, Nature413, 797 (2001)

    work page 2001

  7. [7]

    G. M. Whitesides and B. Grzybowski, Self-assembly at all scales, Science295, 2418 (2002)

    work page 2002

  8. [8]

    K. A. Athanasiou, R. Eswaramoorthy, P. Hadidi, and J. C. Hu, Self-organization and the self-assembling process in tissue engineering, Annual review of biomedical engineering15, 115 (2013)

    work page 2013

  9. [9]

    S. C. Goldstein, J. D. Campbell, and T. C. Mowry, Programmable matter, Computer38, 99 (2005)

    work page 2005

  10. [10]

    Rubenstein, A

    M. Rubenstein, A. Cornejo, and R. Nagpal, Programmable self-assembly in a thousand-robot swarm, Science345, 795 (2014)

    work page 2014

  11. [11]

    Fatehullah, S

    A. Fatehullah, S. H. Tan, and N. Barker, Organoids as an in vitro model of human development and disease, Nature cell biology18, 246 (2016)

    work page 2016

  12. [12]

    Clevers, Modeling development and disease with organoids, Cell165, 1586 (2016)

    H. Clevers, Modeling development and disease with organoids, Cell165, 1586 (2016)

    work page 2016

  13. [13]

    R. S. Sutton, A. G. Barto,et al.,Reinforcement learning: An introduction, Vol. 1 (MIT press Cambridge, 1998)

    work page 1998

  14. [14]

    R. Lowe, Y. I. Wu, A. Tamar, J. Harb, O. Pieter Abbeel, and I. Mordatch, Multi-agent actor-critic for mixed cooperative- competitive environments, Advances in neural information processing systems30(2017)

    work page 2017

  15. [15]

    Baker, I

    B. Baker, I. Kanitscheider, T. Markov, Y. Wu, G. Powell, B. McGrew, and I. Mordatch, Emergent tool use from multi-agent autocurricula, inInternational Conference on Learning Representations (ICLR)(2020)

    work page 2020

  16. [16]

    Pathak, C

    D. Pathak, C. Lu, T. Darrell, P. Isola, and A. A. Efros, Learning to control self-assembling morphologies: a study of generalization via modularity, Advances in Neural Information Processing Systems32(2019)

    work page 2019

  17. [17]

    C. Lin, T. Fan, W. Wang, and M. Nießner, Modeling 3d shapes by reinforcement learning, inComputer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part X 16(Springer, 2020) pp. 545–561

    work page 2020

  18. [18]

    Viquerat, J

    J. Viquerat, J. Rabault, A. Kuhnle, H. Ghraieb, A. Larcher, and E. Hachem, Direct shape optimization through deep reinforcement learning, Journal of Computational Physics428, 110080 (2021)

    work page 2021

  19. [19]

    Deshpande, F

    R. Deshpande, F. Mottes, A.-D. Vlad, M. P. Brenner, and A. Dal Co, Engineering morphogenesis of cell clusters with differentiable programming, Nature Computational Science5, 875 (2025)

    work page 2025

  20. [20]

    Mordvintsev, E

    A. Mordvintsev, E. Randazzo, E. Niklasson, and M. Levin, Growing neural cellular automata, Distill5, e23 (2020)

    work page 2020

  21. [21]

    R. B. Palm, M. Gonz´ alez-Duque, S. Sudhakaran, and S. Risi, Variational neural cellular automata, inProceedings of the 10th International Conference on Learning Representations (ICLR)(2022)

    work page 2022

  22. [22]

    Zhang, C

    D. Zhang, C. Choi, J. Kim, and Y. M. Kim, Learning to generate 3d shapes with generative cellular automata, inProceedings of the International Conference on Learning Representations (ICLR)(2021)

    work page 2021

  23. [23]

    Grattarola, L

    D. Grattarola, L. Livi, and C. Alippi, Learning graph cellular automata, inAdvances in Neural Information Processing Systems, Vol. 34 (2021) pp. 20983–20994

    work page 2021

  24. [24]

    Tesfaldet, D

    M. Tesfaldet, D. Nowrouzezahrai, and C. Pal, Attention-based neural cellular automata, inAdvances in Neural Information Processing Systems, Vol. 35 (Curran Associates, Inc., 2022) pp. 8174–8186

    work page 2022

  25. [25]

    Niu and C

    K. Niu and C. Tian, Zernike polynomials and their applications, Journal of Optics24, 123001 (2022)

    work page 2022

  26. [26]

    Novotni and R

    M. Novotni and R. Klein, Shape retrieval using 3d zernike descriptors, Computer-Aided Design36, 1047 (2004)

    work page 2004

  27. [27]

    Kazhdan, T

    M. Kazhdan, T. Funkhouser, and S. Rusinkiewicz, Rotation invariant spherical harmonic representation of 3 d shape descriptors, inSymposium on geometry processing, Vol. 6 (2003) pp. 156–164

    work page 2003

  28. [28]

    Novotni and R

    M. Novotni and R. Klein, 3d zernike descriptors for content based shape retrieval, inProceedings of the eighth ACM symposium on Solid modeling and applications(2003) pp. 216–225

    work page 2003

  29. [29]

    Scoccimarro, The bispectrum: from theory to observations, The Astrophysical Journal544, 597 (2000)

    R. Scoccimarro, The bispectrum: from theory to observations, The Astrophysical Journal544, 597 (2000)

    work page 2000

  30. [30]

    Kakarala, The bispectrum as a source of phase-sensitive invariants for fourier descriptors: a group-theoretic approach, Journal of Mathematical Imaging and Vision44, 341 (2012)

    R. Kakarala, The bispectrum as a source of phase-sensitive invariants for fourier descriptors: a group-theoretic approach, Journal of Mathematical Imaging and Vision44, 341 (2012)

    work page 2012

  31. [31]

    Collis, P

    W. Collis, P. White, and J. Hammond, Higher-order spectra: the bispectrum and trispectrum, Mechanical systems and signal processing12, 375 (1998)

    work page 1998

  32. [32]

    H. Fan, H. Su, and L. J. Guibas, A point set generation network for 3d object reconstruction from a single image, in Proceedings of the IEEE conference on computer vision and pattern recognition(2017) pp. 605–613. 18

    work page 2017

  33. [33]

    Rubner, C

    Y. Rubner, C. Tomasi, and L. J. Guibas, The earth mover’s distance as a metric for image retrieval, International journal of computer vision40, 99 (2000)

    work page 2000

  34. [34]

    G. Gala, D. Grattarola, and E. Quaeghebeur, E (n)-equivariant graph neural cellular automata, Transactions on Machine Learning Research2024(2024)

    work page 2024

  35. [35]

    M´ emoli, Gromov–wasserstein distances and the metric approach to object matching, Foundations of computational mathematics11, 417 (2011)

    F. M´ emoli, Gromov–wasserstein distances and the metric approach to object matching, Foundations of computational mathematics11, 417 (2011)

    work page 2011

  36. [36]

    V. G. Satorras, E. Hoogeboom, and M. Welling, E (n) equivariant graph neural networks, inInternational conference on machine learning(PMLR, 2021) pp. 9323–9332

    work page 2021

  37. [37]

    Vaswani, N

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, and I. Polosukhin, Attention is all you need, Advances in neural information processing systems30(2017)

    work page 2017

  38. [38]

    R. T. Chen, Y. Rubanova, J. Bettencourt, and D. K. Duvenaud, Neural ordinary differential equations, Advances in neural information processing systems31(2018)

    work page 2018

  39. [39]

    Neural Stochastic Differ- ential Equations: Deep Latent Gaussian Models in the Diffu- sion Limit, 2019

    B. Tzen and M. Raginsky, Neural stochastic differential equations: Deep latent gaussian models in the diffusion limit, arXiv preprint arXiv:1905.09883 (2019)

    work page arXiv 1905

  40. [40]

    Kidger,On Neural Differential Equations, Ph.D

    P. Kidger,On Neural Differential Equations, Ph.D. thesis, University of Oxford (2021)

    work page 2021

  41. [41]

    Cuturi, Sinkhorn distances: Lightspeed computation of optimal transport, Advances in neural information processing systems26(2013)

    M. Cuturi, Sinkhorn distances: Lightspeed computation of optimal transport, Advances in neural information processing systems26(2013)

    work page 2013

  42. [42]

    Peyr´ e, M

    G. Peyr´ e, M. Cuturi,et al., Computational optimal transport: With applications to data science, Foundations and Trends® in Machine Learning11, 355 (2019)

    work page 2019

  43. [43]

    Gonz´ alez, Measurement of areas on a sphere using fibonacci and latitude–longitude lattices, Mathematical Geosciences 42, 49 (2010)

    ´A. Gonz´ alez, Measurement of areas on a sphere using fibonacci and latitude–longitude lattices, Mathematical Geosciences 42, 49 (2010)

    work page 2010

  44. [44]

    Bridson, Fast poisson disk sampling in arbitrary dimensions, inACM SIGGRAPH 2007 sketches(ACM, 2007) p

    R. Bridson, Fast poisson disk sampling in arbitrary dimensions, inACM SIGGRAPH 2007 sketches(ACM, 2007) p. 22

    work page 2007

  45. [45]

    Wartlick, A

    O. Wartlick, A. Kicheva, and M. Gonz´ alez-Gait´ an, Morphogen gradient formation, Cold Spring Harbor perspectives in biology1, a001255 (2009)

    work page 2009

  46. [46]

    Briscoe and S

    J. Briscoe and S. Small, Morphogen rules: design principles of gradient-mediated embryo patterning, Development142, 3996 (2015)

    work page 2015

  47. [47]

    Harland and J

    R. Harland and J. Gerhart, Formation and function of spemann’s organizer, Annual review of cell and developmental biology13, 611 (1997)

    work page 1997

  48. [48]

    E. M. De Robertis, Spemann’s organizer and self-regulation in amphibian embryos, Nature reviews Molecular cell biology 7, 296 (2006)

    work page 2006

  49. [49]

    Jaeger, The gap gene network, Cellular and Molecular Life Sciences68, 243 (2011)

    J. Jaeger, The gap gene network, Cellular and Molecular Life Sciences68, 243 (2011)

    work page 2011

  50. [50]

    A. F. Schier, Nodal morphogens, Cold Spring Harbor perspectives in biology1, a003459 (2009)

    work page 2009

  51. [51]

    E. M. De Robertis and H. Kuroda, Dorsal-ventral patterning and neural induction in xenopus embryos, Annu. Rev. Cell Dev. Biol.20, 285 (2004)

    work page 2004

  52. [52]

    Becht, L

    E. Becht, L. McInnes, J. Healy, C.-A. Dutertre, I. W. Kwok, L. G. Ng, F. Ginhoux, and E. W. Newell, Dimensionality reduction for visualizing single-cell data using umap, Nature biotechnology37, 38 (2019)

    work page 2019

  53. [53]

    Alon,An introduction to systems biology: design principles of biological circuits(Chapman and Hall/CRC, 2019)

    U. Alon,An introduction to systems biology: design principles of biological circuits(Chapman and Hall/CRC, 2019)

    work page 2019

  54. [54]

    E. H. Davidson and M. S. Levine, Properties of developmental gene regulatory networks, Proceedings of the National Academy of Sciences105, 20063 (2008)

    work page 2008

  55. [55]

    Murrell, P

    M. Murrell, P. W. Oakes, M. Lenz, and M. L. Gardel, Forcing cells into shape: the mechanics of actomyosin contractility, Nature reviews Molecular cell biology16, 486 (2015)

    work page 2015

  56. [56]

    B. M. Gumbiner, Regulation of cadherin-mediated adhesion in morphogenesis, Nature reviews Molecular cell biology6, 622 (2005)

    work page 2005

  57. [57]

    Levin, Endogenous bioelectrical networks store non-genetic patterning information during development and regenera- tion, The Journal of physiology592, 2295 (2014)

    M. Levin, Endogenous bioelectrical networks store non-genetic patterning information during development and regenera- tion, The Journal of physiology592, 2295 (2014)

    work page 2014

  58. [58]

    C. H. Waddington,The strategy of the genes(Routledge, 2014)

    work page 2014

  59. [59]

    Wolpert, C

    L. Wolpert, C. Tickle, and A. M. Arias,Principles of development(Oxford University Press, USA, 2015)

    work page 2015

  60. [60]

    Delile, M

    J. Delile, M. Herrmann, N. Peyri´ eras, and R. Doursat, A cell-based computational model of early embryogenesis coupling mechanical behaviour and gene regulation, Nature communications8, 13929 (2017)

    work page 2017

  61. [61]

    C. M. Glen, M. L. Kemp, and E. O. Voit, Agent-based modeling of morphogenetic systems: Advantages and challenges, PLoS computational biology15, e1006577 (2019)

    work page 2019

  62. [62]

    G. Guan, S. Wang, T. G. Shields, S. H. Pahng, C. X. Shao, J. Ye, C. Budjan, and S. Hormoz, Cooperative short-and long-range interactions enable robust symmetry breaking and axis formation, bioRxiv , 2025 (2025)

    work page 2025

  63. [63]

    J. A. Sherratt and J. D. Murray, Models of epidermal wound healing, Proceedings of the Royal Society of London. Series B: Biological Sciences241, 29 (1990)

    work page 1990

  64. [64]

    D. Lobo, W. S. Beane, and M. Levin, Modeling planarian regeneration: a primer for reverse-engineering the worm, PLoS computational biology8, e1002481 (2012)

    work page 2012

  65. [65]

    J. S. Yodh, Y. Lin, S. Sinha, V. Krishnan, L. Mahadevan, and D. J. Cohen, Optimal bioelectric control accelerates collective wound healing, bioRxiv , 2025 (2025)

    work page 2025

  66. [66]

    V. G. Kim, W. Li, N. J. Mitra, S. Chaudhuri, S. DiVerdi, and T. Funkhouser, Learning part-based templates from large collections of 3d shapes, ACM Transactions on Graphics (TOG)32, 1 (2013)

    work page 2013

  67. [67]

    Gray and G

    J. Gray and G. J. Hancock, The propulsion of sea-urchin spermatozoa, Journal of Experimental Biology32, 802 (1955). 19

    work page 1955

  68. [68]

    A. J. Ijspeert, Central pattern generators for locomotion control in animals and robots: a review, Neural networks21, 642 (2008)

    work page 2008

  69. [69]

    D. P. Kingma, Adam: A method for stochastic optimization, arXiv preprint arXiv:1412.6980 (2014)

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  70. [70]

    Hu, Angular trispectrum of the cosmic microwave background, Physical Review D64, 083005 (2001)

    W. Hu, Angular trispectrum of the cosmic microwave background, Physical Review D64, 083005 (2001)

    work page 2001

  71. [71]

    e3nn: Euclidean neural net- works,

    M. Geiger and T. Smidt, e3nn: Euclidean neural networks, arXiv preprint arXiv:2207.09453 (2022)

    work page arXiv 2022

  72. [72]

    Searching for Activation Functions

    P. Ramachandran, B. Zoph, and Q. V. Le, Searching for activation functions, arXiv preprint arXiv:1710.05941 (2017)

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  73. [73]

    McInnes, J

    L. McInnes, J. Healy, N. Saul, and L. Grossberger, Umap: Uniform manifold approximation and projection, The Journal of Open Source Software3, 861 (2018)

    work page 2018

  74. [74]

    V. A. Traag, L. Waltman, and N. J. Van Eck, From louvain to leiden: guaranteeing well-connected communities, Scientific reports9, 1 (2019)

    work page 2019

  75. [75]

    F. A. Wolf, P. Angerer, and F. J. Theis, Scanpy: large-scale single-cell gene expression data analysis, Genome biology19, 15 (2018)

    work page 2018

  76. [76]

    B. C. Hall, Lie groups, lie algebras, and representations, inQuantum Theory for Mathematicians(Springer, 2013) pp. 333–366

    work page 2013

  77. [77]

    Fulton and J

    W. Fulton and J. Harris,Representation theory: a first course, Vol. 129 (Springer Science & Business Media, 2013)

    work page 2013

  78. [78]

    Shoemake, Animating rotation with quaternion curves, inProceedings of the 12th annual conference on Computer graphics and interactive techniques(1985) pp

    K. Shoemake, Animating rotation with quaternion curves, inProceedings of the 12th annual conference on Computer graphics and interactive techniques(1985) pp. 245–254

    work page 1985

  79. [79]

    direct,” dashed line) increases at a rate of 1.7ms per step, whereas implicit differentiation (“Implicit,

    J. You, Z. Ying, and J. Leskovec, Design space for graph neural networks, Advances in Neural Information Processing Systems33, 17009 (2020). 20 Appendix 1: Mathematical Properties of 3D Zernike Polynomials In this section, we briefly summarize the key properties of the 3D Zernike polynomials. For a complete discussion, we refer readers to Niu and Tian [24...

    work page 2020