Learning Without Neurons in Physical Systems

Abstract

Learning is traditionally studied in biological or computational systems. The power of learning frameworks in solving hard inverse problems provides an appealing case for the development of physical learning in which physical systems adopt desirable properties on their own without computational design. It was recently realized that large classes of physical systems can physically learn through local learning rules, autonomously adapting their parameters in response to observed examples of use. We review recent work in the emerging field of physical learning, describing theoretical and experimental advances in areas ranging from molecular self-assembly to flow networks and mechanical materials. Physical learning machines provide multiple practical advantages over computer designed ones, in particular by not requiring an accurate model of the system, and their ability to autonomously adapt to changing needs over time. As theoretical constructs, physical learning machines afford a novel perspective on how physical constraints modify abstract learning theory.

Exploring Emergent Properties in Enzymatic Reaction Networks: Design and Control of Dynamic Functional Systems

“…In section , we have explored the influence of external stimuli to regulate ERN activity, serving as physical learning rules for enhanced task performance. Interestingly, these materials potentially can exhibit task adaptiveness through training with different stimuli . Another important direction for designing life-like materials would be to incorporate some type of fuel regeneration and methods to maintain such materials out of equilibrium as many of the properties associated with living systems (motility, homeostasis, regeneration) require dissipative systems and continuous input of energy.…”

“…Interestingly, these materials potentially can exhibit task adaptiveness through training with different stimuli. 243 Another important direction for designing life-like materials would be to incorporate some type of fuel regeneration and methods to maintain such materials out of equilibrium as many of the properties associated with living systems (motility, homeostasis, regeneration) require dissipative systems and continuous input of energy. In section 5, we have introduced different strategies like energy production, compartmentalization, and communication using ERNs, which facilitate crosstalk between multiscale processes to design truly autonomous systems.…”

Soft matter roadmap*

“…In addition to using machine learning to understand soft matter, one can use soft matter to understand machine learning by developing soft matter systems that solve machine learning tasks by evolving each learning degree of freedom independently according to direct physical influences [407,414,[459][460][461][462][463]] so that the system minimises the desired cost function on its own (figure 34). Such systems, while currently less powerful than computational or biological neural networks, combine the adaptability and parallelisation of the latter with the mathematical simplicity of the former, providing a novel window into how systems can learn [407,464].…”

Qudit machine learning

“…Concurrently, there is the rise of physical neural networks (PNNs), which consist of dedicated physical systems, such as nonlinear optical or mechanical resonators, manipulated to produce specific computational outputs [13][14][15]. To date, PNNs have been explored within classical systems.…”