CoChat
Sign Up
Cited by 27Memory and Neural MechanismsSpatial Cognition and NavigationReinforcement Learning in Robotics

The DOI matches its Crossref record.

Self-Supervised Learning of Representations for Space Generates Multi-Modular Grid Cells

Open access
Schaeffer, Rylan, Khona, Mikail, Ma, Tzuhsuan, Eyzaguirre, Cristobal, Koyejo, Sanmi, Fiete, Ila·2023·Advances in Neural Information Processing Systems 36
DOI 10.52202/075280-1004
Open this paper in CoChat — free accountOpen PDF

Abstract

To solve the spatial problems of mapping, localization and navigation, the mammalian lineage has developed striking spatial representations. One important spatial representation is the Nobel-prize winning grid cells: neurons that represent self-location, a local and aperiodic quantity, with seemingly bizarre non-local and spatially periodic activity patterns of a few discrete periods. Why has the mammalian lineage learnt this peculiar grid representation? Mathematical analysis suggests that this multi-periodic representation has excellent properties as an algebraic code with high capacity and intrinsic error-correction, but to date, there is no satisfactory synthesis of core principles that lead to multi-modular grid cells in deep recurrent neural networks. In this work, we begin by identifying key insights from four families of approaches to answering the grid cell question: coding theory, dynamical systems, function optimization and supervised deep learning. We then leverage our insights to propose a new approach that combines the strengths of all four approaches. Our approach is a self-supervised learning (SSL) framework - including data, data augmentations, loss functions and a network architecture - motivated from a normative perspective, without access to supervised position information or engineering of particular readout representations as needed in previous approaches. We show that multiple grid cell modules can emerge in networks trained on our SSL framework and that the networks and emergent representations generalize well outside their training distribution. This work contains insights for neuroscientists interested in the origins of grid cells as well as machine learning researchers interested in novel SSL frameworks.

Overview

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.52202/075280-1004

Cites

Statistics

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

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?

  • The model based only on grid cells with velocity-dependent recurrent connections requires connections to be instantly tuned by the agent's velocity and is hence unlikely biologically feasible.

    “This requires that neuronal connections are instantly tuned by the agent' velocity and is hence unlikely biologically feasible.”

    From Unfolding the Black Box of Recurrent Neural Networks for Path Integration (2025) · cites this paper as [14]

    Section 5.2 defines the model’s recurrent interaction matrix as a function of the agent’s velocity at each time step. It does not say that neuronal connections must be instantly retuned or conclude that the model is biologically infeasible. (LLM, conf=0.85)

    Experimental — accuracy not yet measured

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

No references could be resolved yet.

Research works printed but not linked: 60

  1. [1][1] Dmitriy Aronov, Rhino Nevers, and David W. Tank. Mapping of a non-spatial dimension by the hippocampal–entorhinal circuit. Nature, 543(7647):719–722, March 2017. Number: 7647 Publisher: Nature Publishing Group.
  2. [2][2] Philip Bachman, R Devon Hjelm, and William Buchwalter. Learning Representations by Maximizing Mutual Information Across Views. In Advances in Neural Information Processing Systems, volume 32. Curran Associates, Inc., 2019.
  3. [3][3] Shahab Bakhtiari, Patrick J. Mineault, Tim Lillicrap, Christopher C. Pack, and Blake Aaron Richards. The functional specialization of visual cortex emerges from training parallel pathways with self-supervised predictive learning. In Advances in Neural Information Processing Systems, October 2021.
  4. [4][4] Randall Balestriero, Mark Ibrahim, Vlad Sobal, Ari Morcos, Shashank Shekhar, Tom Goldstein, Florian Bordes, Adrien Bardes, Gregoire Mialon, Yuandong Tian, et al. A cookbook of self-supervised learning. arXiv preprint arXiv:2304.12210, 2023. · arXiv
  5. [5][5] Andrea Banino, Caswell Barry, Benigno Uria, Charles Blundell, Timothy Lillicrap, Piotr Mirowski, Alexander Pritzel, Martin J. Chadwick, Thomas Degris, Joseph Modayil, Greg Wayne, Hubert Soyer, Fabio Viola, Brian Zhang, Ross Goroshin, Neil Rabinowitz, Razvan Pascanu, Charlie Beattie, Stig Petersen, Amir Sadik, Stephen Gaffney, Helen King, Koray Kavukcuoglu, Demis Hassabis, Raia Hadsell, and Dharshan Kumaran. Vector-based navigation using grid-like representations in artificial agents. Nature, 557(7705):429–433, May 2018.
  6. [6][6] Xiaojun Bao, Eva Gjorgieva, Laura K Shanahan, James D Howard, Thorsten Kahnt, and Jay A Gottfried. Grid-like neural representations support olfactory navigation of a two-dimensional odor space. Neuron, 102(5):1066–1075, 2019.
  7. [7][7] Jacob LS Bellmund, William De Cothi, Tom A Ruiter, Matthias Nau, Caswell Barry, and Christian F Doeller. Deforming the metric of cognitive maps distorts memory. Nature human behaviour, 4(2):177–188, 2020.
  8. [8][8] Charlotte N Boccara, Michele Nardin, Federico Stella, Joseph O'Neill, and Jozsef Csicsvari. The entorhinal cognitive map is attracted to goals. Science, 363(6434):1443–1447, 2019.
  9. [9][9] William N Butler, Kiah Hardcastle, and Lisa M Giocomo. Remembered reward locations restructure entorhinal spatial maps. Science, 363(6434):1447–1452, 2019.
  10. [10][10] Mathilde Caron, Piotr Bojanowski, Julien Mairal, and Armand Joulin. Unsupervised Pre- Training of Image Features on Non-Curated Data. arXiv:1905.01278 [cs], August 2019. arXiv: 1905.01278. · arXiv
  11. [11][11] Mathilde Caron, Ishan Misra, Julien Mairal, Priya Goyal, Piotr Bojanowski, and Armand Joulin. Unsupervised Learning of Visual Features by Contrasting Cluster Assignments. arXiv:2006.09882 [cs], January 2021. arXiv: 2006.09882. · arXiv
  12. [12][12] Rishidev Chaudhuri and Ila Fiete. Computational principles of memory. Nature neuroscience, 19(3):394–403, 2016.
  13. [13][13] Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. A Simple Framework for Contrastive Learning of Visual Representations. arXiv:2002.05709 [cs, stat], June 2020. arXiv: 2002.05709. · arXiv
  14. [14][14] Xinlei Chen and Kaiming He. Exploring Simple Siamese Representation Learning, November 2020. arXiv:2011.10566 [cs]. · arXiv
  15. [15][15] Alexandra O. Constantinescu, Jill X. O'Reilly, and Timothy E. J. Behrens. Organizing concep- tual knowledge in humans with a gridlike code. Science, 352(6292):1464–1468, June 2016. Publisher: American Association for the Advancement of Science.
  16. [16][16] Christopher J Cueva and Xue-Xin Wei. Emergence of grid-like representations by training recurrent neural networks to perform spatial localization. International Conference on Learning Representations, page 19, 2018. 12 http://arxiv.org/abs/2304.12210 http://arxiv.org/abs/1905.01278 http://arxiv.org/abs/2006.09882 http://arxiv.org/abs/2002.05709 http://arxiv.org/abs/2011.10566 · arXiv
  17. [18][18] William Dorrell, Peter E Latham, Timothy EJ Behrens, and James CR Whittington. Actionable neural representations: Grid cells from minimal constraints. International Conference on Learning Representations, 2023.
  18. [19][19] Aleksandr Ermolov, Aliaksandr Siarohin, Enver Sangineto, and Nicu Sebe. Whitening for Self-Supervised Representation Learning. International Conference of Machine Learning, page 10, 2021.
  19. [20][20] Ila R. Fiete, Yoram Burak, and Ted Brookings. What Grid Cells Convey about Rat Location. Journal of Neuroscience, 28(27):6858–6871, July 2008. Publisher: Society for Neuroscience Section: Articles.
  20. [21][21] Ruiqi Gao, Jianwen Xie, Xue-Xin Wei, Song-Chun Zhu, and Ying Nian Wu. On path integration of grid cells: Group representation and isotropic scaling. In M. Ranzato, A. Beygelzimer, Y. Dauphin, P.S. Liang, and J. Wortman Vaughan, editors, Advances in Neural Information Processing Systems, volume 34, pages 28623–28635. Curran Associates, Inc., 2021.
  21. [22][22] Ruiqi Gao, Jianwen Xie, Song-Chun Zhu, and Ying Nian Wu. Learning Grid Cells as Vector Rep- resentation of Self-Position Coupled with Matrix Representation of Self-Motion. International Conference on Learning Representations, page 24, 2019.
  22. [23][23] Richard J. Gardner, Erik Hermansen, Marius Pachitariu, Yoram Burak, Nils A. Baas, Ben- jamin A. Dunn, May-Britt Moser, and Edvard I. Moser. Toroidal topology of population activity in grid cells. Nature, 602(7895):123–128, February 2022. Number: 7895 Publisher: Nature Publishing Group.
  23. [24][24] Peter D Grünwald. The minimum description length principle. MIT press, 2007.
  24. [25][25] Torkel Hafting, Marianne Fyhn, Sturla Molden, May-Britt Moser, and Edvard I. Moser. Mi- crostructure of a spatial map in the entorhinal cortex. Nature, 436(7052):801–806, August 2005. Number: 7052 Publisher: Nature Publishing Group.
  25. [26][26] Kaiming He, Haoqi Fan, Yuxin Wu, Saining Xie, and Ross Girshick. Momentum Contrast for Unsupervised Visual Representation Learning. In 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 9726–9735, Seattle, WA, USA, June 2020. IEEE.
  26. [28][28] Mikail Khona, Sarthak Chandra, and Ila R. Fiete. From smooth cortical gradients to discrete modules: spontaneous and topologically robust emergence of modularity in grid cells. Technical report, bioRxiv, January 2022. Section: New Results Type: article.
  27. [29][29] Mikail Khona, Sarthak Chandra, Joy J Ma, and Ila R Fiete. Winning the lottery with neural connectivity constraints: Faster learning across cognitive tasks with spatially constrained sparse rnns. Neural Computation, pages 1–20, 2023.
  28. [30][30] Mikail Khona and Ila R Fiete. Attractor and integrator networks in the brain. Nature Reviews Neuroscience, pages 1–23, 2022.
  29. [31][31] Talia Konkle and George A Alvarez. A self-supervised domain-general learning framework for human ventral stream representation. Nature communications, 13(1):1–12, 2022.
  30. [32][32] Julija Krupic, Marius Bauza, Stephen Burton, Caswell Barry, and John O'Keefe. Grid cell symmetry is shaped by environmental geometry. Nature, 518(7538):232–235, 2015.
  31. [33][33] Julija Krupic, Marius Bauza, Stephen Burton, and John O'Keefe. Framing the grid: effect of boundaries on grid cells and navigation. The Journal of physiology, 594(22):6489–6499, 2016.
  32. [34][34] Julija Krupic, Marius Bauza, Stephen Burton, and John O'Keefe. Local transformations of the hippocampal cognitive map. Science, 359(6380):1143–1146, 2018. 13
  33. [35][35] Xin Liu, Zhongdao Wang, Ya-Li Li, and Shengjin Wang. Self-supervised learning via maximum entropy coding. Advances in Neural Information Processing Systems, 35:34091–34105, 2022.
  34. [36][36] Ziming Liu, Mikail Khona, Ila R Fiete, and Max Tegmark. Growing brains: Co-emergence of anatomical and functional modularity in recurrent neural networks. arXiv preprint arXiv:2310.07711, 2023. · arXiv
  35. [37][37] Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101, 2017. · arXiv
  36. [38][38] Tzuhsuan Ma and Ila R Fiete. Towards a theory for the emergence of grid and place cell codes. PhD thesis, Massachusetts Institute of Technology, 2020.
  37. [39][39] Alexander Mathis, Andreas V. M. Herz, and Martin Stemmler. Optimal Population Codes for Space: Grid Cells Outperform Place Cells. Neural Computation, 24(9):2280–2317, September 2012.
  38. [40][40] Patrick Mineault, Shahab Bakhtiari, Blake Richards, and Christopher Pack. Your head is there to move you around: Goal-driven models of the primate dorsal pathway. In Advances in Neural Information Processing Systems, volume 34, pages 28757–28771. Curran Associates, Inc., 2021.
  39. [42][42] Serdar Ozsoy, Shadi Hamdan, Sercan Arik, Deniz Yuret, and Alper Erdogan. Self-supervised learning with an information maximization criterion. In S. Koyejo, S. Mohamed, A. Agarwal, D. Belgrave, K. Cho, and A. Oh, editors, Advances in Neural Information Processing Systems, volume 35, pages 35240–35253. Curran Associates, Inc., 2022.
  40. [43][43] Seongmin A Park, Douglas S Miller, and Erie D Boorman. Inferences on a multidimensional social hierarchy use a grid-like code. Nature neuroscience, 24(9):1292–1301, 2021.
  41. [44][44] Seongmin A Park, Douglas S Miller, Hamed Nili, Charan Ranganath, and Erie D Boorman. Map making: constructing, combining, and inferring on abstract cognitive maps. Neuron, 107(6):1226–1238, 2020.
  42. [45][45] Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, et al. Pytorch: An imperative style, high-performance deep learning library. Advances in neural information processing systems, 32, 2019.
  43. [47][47] Rylan Schaeffer, Mikail Khona, and Ila R Fiete. No free lunch from deep learning in neu- roscience: A case study through models of the entorhinal-hippocampal circuit. In Alice H. Oh, Alekh Agarwal, Danielle Belgrave, and Kyunghyun Cho, editors, Advances in Neural Information Processing Systems, 2022.
  44. [48][48] Rylan Schaeffer, Mikail Khona, and Ila Rani Fiete. No Free Lunch from Deep Learning in Neuroscience: A Case Study through Models of the Entorhinal-Hippocampal Circuit. Neural Information Processing Systems, August 2022.
  45. [49][49] Rylan Schaeffer, Mikail Khona, Leenoy Meshulam, Brain Laboratory International, and Ila Fiete. Reverse-engineering recurrent neural network solutions to a hierarchical inference task for mice. Advances in Neural Information Processing Systems, 33:4584–4596, 2020.
  46. [50][50] Vemund Schøyen, Markus Borud Pettersen, Konstantin Holzhausen, Marianne Fyhn, Anders Malthe-Sørenssen, and Mikkel Elle Lepperød. Coherently remapping toroidal cells but not grid cells are responsible for path integration in virtual agents. bioRxiv, pages 2022–08, 2022.
  47. [51][51] Ravid Shwartz-Ziv, Randall Balestriero, Kenji Kawaguchi, Tim G. J. Rudner, and Yann LeCun. An information-theoretic perspective on variance-invariance-covariance regularization, 2023. 14 http://arxiv.org/abs/2310.07711 http://arxiv.org/abs/1711.05101 · arXiv
  48. [53][53] Ben Sorscher, Gabriel C Mel, Surya Ganguli, and Samuel A Ocko. A unified theory for the origin of grid cells through the lens of pattern formation. Advances in Neural Information Processing Systems, page 18, 2019.
  49. [54][54] Ben Sorscher, Gabriel C Mel, Samuel A Ocko, Lisa M Giocomo, and Surya Ganguli. A unified theory for the computational and mechanistic origins of grid cells. Neuron, 111(1):121–137, 2023.
  50. [55][55] Sameet Sreenivasan and Ila Fiete. Grid cells generate an analog error-correcting code for singularly precise neural computation. Nature Neuroscience, 14(10):1330–1337, October 2011. Number: 10 Publisher: Nature Publishing Group.
  51. [56][56] Kimberly L Stachenfeld, Matthew M Botvinick, and Samuel J Gershman. The hippocampus as a predictive map. Nature neuroscience, 20(11):1643–1653, 2017.
  52. [57][57] Martin Stemmler, Alexander Mathis, and Andreas VM Herz. Connecting multiple spatial scales to decode the population activity of grid cells. Science Advances, 1(11):e1500816, 2015.
  53. [58][58] Hanne Stensola, Tor Stensola, Trygve Solstad, Kristian Frøland, May-Britt Moser, and Edvard I. Moser. The entorhinal grid map is discretized. Nature, 492(7427):72–78, December 2012. Number: 7427 Publisher: Nature Publishing Group.
  54. [59][59] Joshua B Tenenbaum, Vin de Silva, and John C Langford. A global geometric framework for nonlinear dimensionality reduction. science, 290(5500):2319–2323, 2000.
  55. [60][60] Yonglong Tian, Dilip Krishnan, and Phillip Isola. Contrastive Representation Distillation. arXiv:1910.10699 [cs, stat], January 2020. arXiv: 1910.10699. · arXiv
  56. [64][64] James C. R. Whittington, Timothy H. Muller, Shirley Mark, Guifen Chen, Caswell Barry, Neil Burgess, and Timothy E. J. Behrens. The Tolman-Eichenbaum Machine: Unifying Space and Relational Memory through Generalization in the Hippocampal Formation. Cell, 183(5):1249– 1263.e23, November 2020.
  57. [66][66] Dehong Xu, Ruiqi Gao, Wen-Hao Zhang, Xue-Xin Wei, and Ying Nian Wu. Conformal isometry of lie group representation in recurrent network of grid cells. arXiv preprint arXiv:2210.02684, 2022. · arXiv
  58. [67][67] KiJung Yoon, Michael A Buice, Caswell Barry, Robin Hayman, Neil Burgess, and Ila R Fiete. Specific evidence of low-dimensional continuous attractor dynamics in grid cells. Nature neuroscience, 16(8):1077–1084, 2013.
  59. [68][68] Jure Zbontar, Li Jing, Ishan Misra, Yann LeCun, and Stéphane Deny. Barlow Twins: Self- Supervised Learning via Redundancy Reduction, June 2021. arXiv:2103.03230 [cs, q-bio]. · arXiv
  60. [69][69] Chengxu Zhuang, Siming Yan, Aran Nayebi, Martin Schrimpf, Michael C. Frank, James J. DiCarlo, and Daniel L. K. Yamins. Unsupervised neural network models of the ventral visual stream. Proceedings of the National Academy of Sciences, 118(3):e2014196118, January 2021. 15 http://arxiv.org/abs/1910.10699 http://arxiv.org/abs/2210.02684 http://arxiv.org/abs/2103.03230 A Experimental Details Architecture and training data + augmentations Our code was implemented in PyTorch [45] and PyTorch Lightning. Hyperparameters for our experiments are listed in Table 1. Our code will be made publicly available upon publication. Hyperparameters Values Batch size 130 Trajectory length 60 Velocity sampling distribution vt ∈ R2 ∼i.i.d. Uniform2(−0.15, 0.15) meters RNN nonlinearity Norm(ReLU(·)) Number of RNN units 128 Number of MLP layers 3 Spatial length scale σx 0.05 meters Neural length scale σg 0.4 Separation loss coefficient λSep 1.0 Invariance loss coefficient λInv 0.1 Capacity loss coefficient λCap 0.5 Optimizer AdamW [37] Optimizer scheduler Reduce Learning Rate on Plateau Learning rate 2e-5 Gradient clip value 0.1 Weight decay None Accumulate gradient batches 2 Number of gradient descent steps 2e6 Table 1: Hyperparameters used for training the networks. 16 https://pytorch.org/docs/stable/generated/torch.optim.lr_scheduler.ReduceLROnPlateau.html B All Ratemaps 17 a b c Figure 8: All 128 ratemaps evaluated on trajectories inside a 2m box. The red ratemaps are dead ReLU units. (a) Ratemaps from the corresponding to Fig. 4 (b) Ratemaps corresponding to an ablation of the conformal isometry loss. Notice the existence of multiple modules but the lattice structure is more sheared, closer to being a square than a hexagon. (c) Ratemaps corresponding to the run analyzed in Fig.6. 18 Φ2 Φ1 R0 RI a λ λ1 2 x Φ = 0 Φ = π Φ = 0 Φ = π b Figure 9: (a) 2 Example cells from 2 modules, with preferred phase 0 and π. (b) Visualizing the state space defined by {ϕ1, ϕ2} as a torus (left) and on a square with periodic boundary conditions (right), which an equivalent construction of a torus. C Construction of the grid code To explain the structure of the grid code, we consider idealized tuning curves in 1d. Each cell i is defined by its periodicity λα ∈ R and preferred phase ϕi ∈ S1. All cells in the same module α have the same periodicity and uniformly tile all allowed phases. For position x ∈ R, rαi (x) = RmaxReLU [ cos ( 2π λα x+ ϕi )] · arXiv

Cited by

Sorted by citations

Metadata

Authors
Schaeffer, R., Khona, M., Ma, T., Eyzaguirre, C., Koyejo, S., & Fiete, I.
ORCID
Venue
Advances in Neural Information Processing Systems 36
Publisher
Neural Information Processing Systems Foundation, Inc. (NeurIPS)
Published
2023
Pages
23140-23157
ISSN
1049-5258 (ISSN-L)
DOI
10.52202/075280-1004
OpenAlex
W7133224655
Cited by
27
References
0
Citations by year
2026: 3 · 2025: 10 · 2024: 13 · 2023: 1
FWCI
13.90
Type
conference paper
Language
en
Open access
Licence
cc-by-nc-nd
Locations
  • neural information processing systems (publishedVersion)
  • arXiv (Cornell University) (submittedVersion, cc-by-nc-nd)
  • arXiv (Cornell University)
Topics
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Memory and Neural Mechanisms
  • Social Sciences › Psychology › Experimental and Cognitive Psychology › Spatial Cognition and Navigation
  • Physical Sciences › Computer Science › Artificial Intelligence › Reinforcement Learning in Robotics
Keywords
self-supervised learning, representation learning, grid cells, space representation, spatial navigation, neural network model

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