- Physical Review Journals
All JournalsPhysics Magazine

Physical Review Research

Reuse & Permissions

It is not necessary to obtain permission to reuse this article or its components as it is available under the terms of the Creative Commons Attribution 4.0 International license. This license permits unrestricted use, distribution, and reproduction in any medium, provided attribution to the author(s) and the published article's title, journal citation, and DOI are maintained. Please note that some figures may have been included with permission from other third parties. It is your responsibility to obtain the proper permission from the rights holder directly for these figures.

Export citation

Export citation

Choose format for download:

Download Citation
  • Letter
  • Open Access

Physical learning beyond the quasistatic limit

Menachem Stern1,*, Sam Dillavou1, Marc Z. Miskin2, Douglas J. Durian1, and Andrea J. Liu1

  • 1Department of Physics and Astronomy, University of Pennsylvania, Philadelphia, Pennsylvania 19104, USA
  • 2Department of Electrical and Systems Engineering, University of Pennsylvania, Philadelphia, Pennsylvania 19104, USA

  • *Corresponding author: nachis@sas.upenn.edu
PDF

Phys. Rev. Research 4, L022037 – Published 16 May, 2022

DOI: https://doi.org/10.1103/PhysRevResearch.4.L022037

Abstract

Physical networks, such as biological neural networks, can learn desired functions without a central processor, using local learning rules in space and time to learn in a fully distributed manner. Learning approaches such as equilibrium propagation, directed aging, and coupled learning similarly exploit local rules to accomplish learning in physical networks such as mechanical, flow, or electrical networks. In contrast to certain natural neural networks, however, such approaches have so far been restricted to the quasistatic limit, where they learn on timescales slow compared to their physical relaxation. This quasistatic constraint slows down learning, limiting the use of these methods as machine learning algorithms, and potentially restricting physical networks that could be used as learning platforms. Here we explore learning in an electrical resistor network that implements coupled learning, both in the laboratory and on the computer, at rates that range from slow to far above the quasistatic limit. We find that up to a critical threshold in the ratio of the learning rate to the physical rate of relaxation, learning speeds up without much change of behavior or error. Beyond the critical threshold, the error exhibits oscillatory dynamics but the networks still learn successfully.

View figure in article

Physics Subject Headings (PhySH)

Article Text

Supplemental Material

References (31)

  1. J. R. Movellan, Contrastive Hebbian learning in the continuous Hopfield model, in Connectionist Models (Elsevier, San Diego, USA, 1991), pp. 10–17.
  2. V. Lopez-Pastor and F. Marquardt, Self-learning machines based on Hamiltonian echo backpropagation, arXiv:2103.04992.
  3. E. Martin, M. Ernoult, J. Laydevant, S. Li, D. Querlioz, T. Petrisor, and J. Grollier, EqSpike: Spike-driven equilibrium propagation for neuromorphic implementations, iScience 24, 102222 (2021).
  4. N. Pashine, Local rules for fabricating allosteric networks, Phys. Rev. Materials 5, 065607 (2021).
  5. B. Scellier and Y. Bengio, Equilibrium propagation: Bridging the gap between energy-based models and backpropagation, Front. Comput. Neurosci. 11, 24 (2017).
  6. J. Kendall, R. Pantone, K. Manickavasagam, Y. Bengio, and B. Scellier, Training end-to-end analog neural networks with equilibrium propagation, arXiv:2006.01981.
  7. N. Pashine, D. Hexner, A. J. Liu, and S. R. Nagel, Directed aging, memory, and natures greed, Sci. Adv. 5, eaax4215 (2019).
  8. D. Hexner, N. Pashine, A. J. Liu, and S. R. Nagel, Effect of directed aging on nonlinear elasticity and memory formation in a material, Phys. Rev. Research 2, 043231 (2020).
  9. D. Hexner, A. J. Liu, and S. R. Nagel, Periodic training of creeping solids, Proc. Natl. Acad. Sci. USA 117, 31690 (2020).
  10. M. Stern, V. Jayaram, and A. Murugan, Shaping the topology of folding pathways in mechanical systems, Nat. Commun. 9, 4303 (2018).
  11. M. Stern, M. B. Pinson, and A. Murugan, Continual Learning of Multiple Memories in Mechanical Networks, Phys. Rev. X 10, 031044 (2020).
  12. M. Stern, C. Arinze, L. Perez, S. E. Palmer, and A. Murugan, Supervised learning through physical changes in a mechanical system, Proc. Natl. Acad. Sci. USA 117, 14843 (2020).
  13. M. Stern, D. Hexner, J. W. Rocks, and A. J. Liu, Supervised Learning in Physical Networks: From Machine Learning to Learning Machines, Phys. Rev. X 11, 021045 (2021).
  14. S. Dillavou, M. Stern, A. J. Liu, and D. J. Durian, Demonstration of decentralized, physics-driven learning, arXiv:2108.00275.
  15. J. F. Wycoff, S. Dillavou, M. Stern, A. J. Liu, and D. J. Durian, Desynchronous learning in a physics-driven learning network, J. Chem. Phys. 156, 144903 (2022).
  16. S. Bartunov, A. Santoro, B. Richards, L. Marris, G. E. Hinton, and T. Lillicrap, Assessing the scalability of biologically-motivated deep learning algorithms and architectures, in Advances in Neural Information Processing Systems (NeurIPS, Montreal, Canada, 2018), pp. 9368–9378.
  17. M. Ernoult, J. Grollier, D. Querlioz, Y. Bengio, and B. Scellier, Updates of equilibrium prop match gradients of backprop through time in an RNN with static input, Advances in Neural Information Processing Systems 32 (NeurIPS, Montreal, Canada, 2019).
  18. M. Ernoult, J. Grollier, D. Querlioz, Y. Bengio, and B. Scellier, Equilibrium propagation with continual weight updates, arXiv:2005.04168.
  19. B. Scellier, A deep learning theory for neural networks grounded in physics, arXiv:2103.09985.
  20. R. S. Zucker and W. G. Regehr, Short-term synaptic plasticity, Annu. Rev. Physiol. 64, 355 (2002).
  21. R. Kanai and F. A. J. Verstraten, Perceptual manifestations of fast neural plasticity: Motion priming, rapid motion aftereffect and perceptual sensitization, Vis. Res. 45, 3109 (2005).
  22. S. Marom, Neural timescales or lack thereof, Prog. Neurobiol. 90, 16 (2010).
  23. Y. Sagi, I. Tavor, S. Hofstetter, S. Tzur-Moryosef, T. Blumenfeld-Katzir, and Y. Assaf, Learning in the fast lane: New insights into neuroplasticity, Neuron 73, 1195 (2012).
  24. S. M. Lemon and L. N. Binn, Serum neutralizing antibody response to hepatitis A virus, J. Infect. Dis. 148, 1033 (1983).
  25. WHO Guidelines on Hepatitis B and C Testing (World Health Organization Press, Geneve, Switzerland, 2017).
  26. J. E. Banatvala and D. W. G. Brown, Rubella, Lancet 363, 1127 (2004).
  27. The update rule is slightly different, now consisting of the evaluation of XOR(ΔvjF>ΔvjC,ΔvjF>−ΔvjC) whereas before it was XOR(ΔvjF>ΔvjC,ΔvjC>0). We have observed no effect from this change, but implemented it to adhere more closely to the original coupled learning rule.
  28. See Supplemental Material at http://link.aps.org/supplemental/10.1103/PhysRevResearch.4.L022037 for more details on the experiments, simulations, and accompanying theory.
  29. J. W. Rocks, H. Ronellenfitsch, A. J. Liu, S. R. Nagel, and E. Katifori, Limits of multifunctionality in tunable networks, Proc. Natl. Acad. Sci. USA 116, 2506 (2019).
  30. M. Ruiz-García, A. J. Liu, and E. Katifori, Tuning and jamming reduced to their minima, Phys. Rev. E 100, 052608 (2019).
  31. M. Ruiz-Garcia, G. Zhang, S. S. Schoenholz, and A. J. Liu, Tilting the playing field: Dynamical loss functions for machine learning, in International Conference on Machine Learning (PMLR, (ICML 2021) virtual conference, 2021), 9157–9167.

Outline

Information

Phys. Rev. Research 4, L022037– Published 16 May, 2022
  • Received 10 January 2022
  • Accepted 28 April 2022
Crossmark - Check for updates button
Creative Commons Logo

Published by the American Physical Society under the terms of the Creative Commons Attribution 4.0 International license. Further distribution of this work must maintain attribution to the author(s) and the published article's title, journal citation, and DOI.

Published by the American Physical Society

Sign In to Your Journals Account

Filter

Filter

Article Lookup

Enter a citation