Condensed Matter > Disordered Systems and Neural Networks
[Submitted on 29 Dec 2023]
Title:Contrastive learning through non-equilibrium memory
View PDF HTML (experimental)Abstract:Learning algorithms based on backpropagation have enabled transformative technological advances but alternatives based on local energy-based rules offer benefits in terms of biological plausibility and decentralized training. A broad class of such local learning rules involve \textit{contrasting} a clamped configuration with the free, spontaneous behavior of the system. However, comparisons of clamped and free configurations require explicit memory or switching between Hebbian and anti-Hebbian modes. Here, we show how a simple form of implicit non-equilibrium memory in the update dynamics of each ``synapse'' of a network naturally allows for contrastive learning. During training, free and clamped behaviors are shown in sequence over time using a sawtooth-like temporal protocol that breaks the symmetry between those two behaviors when combined with non-equilibrium update dynamics at each synapse. We show that the needed dynamics is implicit in integral feedback control, broadening the range of physical and biological systems naturally capable of contrastive learning. Finally, we show that non-equilibrium dissipation improves learning quality and determine the Landauer energy cost of contrastive learning through physical dynamics.
Current browse context:
cond-mat.dis-nn
Change to browse by:
References & Citations
Bibliographic and Citation Tools
Bibliographic Explorer (What is the Explorer?)
Litmaps (What is Litmaps?)
scite Smart Citations (What are Smart Citations?)
Code, Data and Media Associated with this Article
CatalyzeX Code Finder for Papers (What is CatalyzeX?)
DagsHub (What is DagsHub?)
Gotit.pub (What is GotitPub?)
Papers with Code (What is Papers with Code?)
ScienceCast (What is ScienceCast?)
Demos
Recommenders and Search Tools
Influence Flower (What are Influence Flowers?)
Connected Papers (What is Connected Papers?)
CORE Recommender (What is CORE?)
IArxiv Recommender
(What is IArxiv?)
arXivLabs: experimental projects with community collaborators
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.