The Physical Effects of Learning

Abstract

Interacting many-body physical systems ranging from neural networks in the brain to folding proteins to self-modifying electrical circuits can learn to perform diverse tasks. This learning, both in nature and in engineered systems, can occur through evolutionary selection or through dynamical rules that drive active learning from experience. Here, we show that \added{learning in linear physical networks with weak input signals} leaves architectural imprints on the Hessian of a physical system. Compared to a generic organization of the system components, (a) the effective physical dimension of the response to inputs decreases, (b) the response of physical degrees of freedom to random perturbations (or system ``susceptibility'') increases, and (c) the low-eigenvalue eigenvectors of the Hessian align with the task. Overall, these effects embody the typical scenario for learning processes in physical systems in the weak input regime, suggesting ways of discovering whether a physical network may have been trained.

arXiv comment24 pages, 10 figures

References in corpus (22)

• A high-bias, low-variance introduction to Machine Learning for physicists
• Designing allostery-inspired response in mechanical networks
• Anharmonicity and quasi-localization of the excess low-frequency vibrations in jammed solids
• Network Plasticity as Bayesian Inference
• Architecture and Co-Evolution of Allosteric Materials
• Learning without neurons in physical systems
• Directed aging, memory and Nature's greed
• Demonstration of Decentralized, Physics-Driven Learning
• Memorizing without overfitting: Bias, variance, and interpolation in over-parameterized models
• The limits of multifunctionality in tunable networks
• Supervised learning in physical networks: From machine learning to learning machines
• Physical model of the genotype-to-phenotype map of proteins
• Shaping dynamical folding and misfolding pathways in mechanical metamaterials
• Principles for optimal cooperativity in allosteric materials
• Effect of aging on the non-linear elasticity and memory formation in materials
• Desynchronous Learning in a Physics-Driven Learning Network
• Self-learning Machines based on Hamiltonian Echo Backpropagation
• Learned multi-stability in mechanical networks
• Local rules for fabricating allosteric networks
• Mechanics of allostery: contrasting the induced fit and population shift scenarios
• Learning to self-fold at a bifurcation
• Tuning and jamming reduced to their minima

Cited by in corpus (7)

• Machine Learning Without a Processor: Emergent Learning in a Nonlinear Electronic Metamaterial
• Rigidity of Epithelial Tissues as a Double Optimization Problem
• Self-organization and memory in an disordered solid subject to random loading
• Physical networks become what they learn
• Equilibrium Propagation: the Quantum and the Thermal Cases
• A universal approximation theorem for nonlinear resistive networks
• Harnessing intuitive local evolution rules for physical learning