Temporal Contrastive Learning through implicit non-equilibrium memory

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Falk, Martin J., Strupp, Adam T., Scellier, Benjamin, Murugan, Arvind
Format: Preprint
Published: 2023
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912207851225088
author Falk, Martin J.
Strupp, Adam T.
Scellier, Benjamin
Murugan, Arvind
author_facet Falk, Martin J.
Strupp, Adam T.
Scellier, Benjamin
Murugan, Arvind
contents The backpropagation method has enabled transformative uses of neural networks. Alternatively, for energy-based models, local learning methods involving only nearby neurons offer benefits in terms of decentralized training, and allow for the possibility of learning in computationally-constrained substrates. One class of local learning methods constrasts the desired, clamped behavior with spontaneous, free behavior. However, directly contrasting free and clamped behaviors requires explicit memory. Here, we introduce `Temporal Contrastive Learning', an approach that uses integral feedback in each learning degree of freedom to provide a simple form of implicit non-equilibrium memory. During training, free and clamped behaviors are shown in a sawtooth-like protocol over time. When combined with integral feedback dynamics, these alternating temporal protocols generate an implicit memory necessary for comparing free and clamped behaviors, broadening the range of physical and biological systems capable of contrastive learning. Finally, we show that non-equilibrium dissipation improves learning quality and determine a Landauer-like energy cost of contrastive learning through physical dynamics.
format Preprint
id arxiv_https___arxiv_org_abs_2312_17723
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Temporal Contrastive Learning through implicit non-equilibrium memory
Falk, Martin J.
Strupp, Adam T.
Scellier, Benjamin
Murugan, Arvind
Disordered Systems and Neural Networks
Soft Condensed Matter
Statistical Mechanics
Molecular Networks
The backpropagation method has enabled transformative uses of neural networks. Alternatively, for energy-based models, local learning methods involving only nearby neurons offer benefits in terms of decentralized training, and allow for the possibility of learning in computationally-constrained substrates. One class of local learning methods constrasts the desired, clamped behavior with spontaneous, free behavior. However, directly contrasting free and clamped behaviors requires explicit memory. Here, we introduce `Temporal Contrastive Learning', an approach that uses integral feedback in each learning degree of freedom to provide a simple form of implicit non-equilibrium memory. During training, free and clamped behaviors are shown in a sawtooth-like protocol over time. When combined with integral feedback dynamics, these alternating temporal protocols generate an implicit memory necessary for comparing free and clamped behaviors, broadening the range of physical and biological systems capable of contrastive learning. Finally, we show that non-equilibrium dissipation improves learning quality and determine a Landauer-like energy cost of contrastive learning through physical dynamics.
title Temporal Contrastive Learning through implicit non-equilibrium memory
topic Disordered Systems and Neural Networks
Soft Condensed Matter
Statistical Mechanics
Molecular Networks
url https://arxiv.org/abs/2312.17723