Stochastic Thermodynamics of Learning

Abstract

Virtually every organism gathers information about its noisy environment and builds models from that data, mostly using neural networks. Here, we use stochastic thermodynamics to analyse the learning of a classification rule by a neural network. We show that the information acquired by the network is bounded by the thermodynamic cost of learning and introduce a learning efficiency $\eta \le 1$. We discuss the conditions for optimal learning and analyse Hebbian learning in the thermodynamic limit.

arXiv comment5 pages, 3 figures, 7 pages of supplemental material

References in corpus (19)

• Stochastic thermodynamics, fluctuation theorems, and molecular machines
• Thermodynamic uncertainty relation for biomolecular processes
• Generalized Jarzynski Equality under Nonequilibrium Feedback Control
• The Energetic Costs of Cellular Computation
• Thermodynamics with continuous information flow
• Stochastic thermodynamics of bipartite systems: transfer entropy inequalities and a Maxwell's demon interpretation
• Extracting work from a single heat bath through feedback
• Thermodynamic costs of information processing in sensory adaption
• Efficiency of cellular information processing
• Maxwell's demon in biochemical signal transduction with feedback loop
• Second-law-like inequalities with information and their interpretations
• Optimizing information flow in small genetic networks. I
• Thermodynamics of statistical inference by cells
• Thermodynamic efficiency of information and heat flow
• Sensory capacity: an information theoretical measure of the performance of a sensor
• Multipartite information flow for multiple Maxwell demons
• Nonequilibrium sensing and its analogy to kinetic proofreading
• Unraveling Adaptation in Eukaryotic Pathways: Lessons from Protocells
• Kinetics and thermodynamics of reversible polymerization in closed systems

Cited by in corpus (23)

• Stochastic thermodynamics of computation
• Uncertainty relations in stochastic processes: An information inequality approach
• Stochastic thermodynamic interpretation of information geometry
• Deep Learning Theory Review: An Optimal Control and Dynamical Systems Perspective
• Thermodynamic Computing via Autonomous Quantum Thermal Machines
• Thermodynamic uncertainty relation of interacting oscillators in synchrony
• Thermodynamic uncertainty relation for time-delayed Langevin systems
• Uncertainty relations for time-delayed Langevin systems
• Nonequilibrium Thermodynamics of Restricted Boltzmann Machines
• Quantifying Relevance in Learning and Inference
• What we learn from the learning rate
• Effective thermodynamics of two interacting underdamped Brownian particles
• A Physical Quantum Agent
• Stochastic thermodynamics of an electron-spin-resonance quantum dot system
• Thermodynamic efficiency of learning a rule in neural networks
• Information thermodynamics: from physics to neuroscience
• Thermodynamic role of main reaction pathway and multi-body information flow in membrane transport
• Thermodynamic fluctuation theorems govern human sensorimotor learning
• Hidden energy flows in strongly coupled nonequilibrium systems
• Jarzyski's equality and Crooks' fluctuation theorem for general Markov chains with application to decision-making systems
• Dissipation in non-steady state regulatory circuits
• Nonequilibrium thermodynamics of self-supervised learning
• Thermodynamic Limits of Physical Intelligence