Desynchronous learning in a physics-driven learning network

Wycoff, J. F.; Dillavou, Sam; Stern, Menachem; Liu, Andrea J.; Durian, Douglas J.

doi:10.1063/5.0084631

Cited by 19 publications

(16 citation statements)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Early experimental systems implementing contrastive learning in linear resistor and elastic networks (Fig. 3F,G) demonstrated the success of this approach (92,27,28), e.g. in classifying the Iris dataset (Fig.…”

Section: Contrastive Learningmentioning

confidence: 97%

“…Nevertheless, biologically plausible learning rules, simulated on computers, have proven successful (25), suggesting that physical learning has potential despite locality constraints (26). Further, locality provides benefits as well, since learning can be desynchronous, more robust and scale better with system size since it does not rely on a central processor (27,28).…”

Section: Challenge and Opportunity: Local Learning Rulesmentioning

confidence: 99%

See 1 more Smart Citation

Learning without neurons in physical systems

Stern¹,

Murugan²

2022

Preprint

View full text Add to dashboard Cite

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.

show abstract

Section: Contrastive Learningmentioning

confidence: 97%

Section: Challenge and Opportunity: Local Learning Rulesmentioning

confidence: 99%

Learning without neurons in physical systems

Stern¹,

Murugan²

2022

Preprint

View full text Add to dashboard Cite

show abstract

“…Such mechanical networks are capable of retaining multiple memories, such as multiple allosteric responses. Analogous approaches carry over to flow and electrical networks [417,419], with corresponding physical cost functions that the system may optimise, as well as (design) cost functions which represent the design goals. The ability (a) (Physics-driven learning network) A network of resistors is allowed to evolve by changing values of the resistors based on local rules so that it can generate specified voltages at a set of target nodes (blue boxes), given input voltages at input nodes (red circles).…”

Section: Current and Future Challengesmentioning

confidence: 99%

Soft matter roadmap^*

Barrat,

Del Gado,

Egelhaaf

et al. 2023

J. Phys. Mater.

View full text Add to dashboard Cite

Soft materials are usually defined as materials made of mesoscopic entities, often self-organized, sensitive to thermal fluctuations and to weak perturbations. Archetypal examples are colloids, polymers, amphiphiles, liquid crystals, foams. The importance of soft materials in everyday commodity products, as well as in technological applications, is enormous, and controlling or improving their properties is the focus of many efforts. From a fundamental perspective, the possibility of manipulating soft material properties, by tuning interactions between constituents and by applying external perturbations, gives rise to an almost unlimited variety in physical properties. Together with the relative ease to observe and characterize them, this renders soft matter systems powerful model systems to investigate statistical physics phenomena, many of them relevant as well to hard condensed matter systems. Understanding the emerging properties from mesoscale constituents still poses enormous challenges, which have stimulated a wealth of new experimental approaches, including the synthesis of new systems with, e.g., tailored self-assembling properties, or novel experimental techniques in imaging, scattering or rheology. Theoretical and numerical methods, and coarse-grained models, have become central to predict physical properties of soft materials, while computational approaches that also use machine learning tools are playing a progressively major role in many investigations.This roadmap paper intends to give a broad overview of recent and possible future activities in the field of soft materials, with experts covering various developments and challenges in material synthesis and characterization, instrumental, simulation and theoretical methods as well as general concepts.

show abstract

“…In previous work, we constructed a basic version of such a system, 11 and used it to investigate various modes of physical learning. 12,13…”

Section: Introductionmentioning

confidence: 99%

Circuits that train themselves: decentralized, physics-driven learning

Dillavou

Beyer

Stern

et al. 2023

AI and Optical Data Sciences IV

View full text Add to dashboard Cite

In typical artificial neural networks, neurons adjust according to global calculations of a central processor, but in the brain neurons and synapses self-adjust based on local information. A man-made self-adjusting (distributed) system capable of performing machine-learning problems would have substantial scaling advantages over typical computational neural networks, in power consumption, speed, and robustness to damage. Furthermore, such a system would allow us to study physical learning without the added complexity of biology. Here we unveil the second-generation design of such a system -a transistor-based self-adjusting analog network that trains itself to perform a wide variety of tasks. Here we demonstrate basic features of the system, including the ability to monitor all internal states. This platform is already faster than a simulation of itself, and is thus an exciting platform for the investigation of physical learning.

show abstract

Desynchronous learning in a physics-driven learning network

Cited by 19 publications

References 22 publications

Learning without neurons in physical systems

Learning without neurons in physical systems

Soft matter roadmap^*

Circuits that train themselves: decentralized, physics-driven learning

Contact Info

Product

Resources

About

Desynchronous learning in a physics-driven learning network

Cited by 19 publications

References 22 publications

Learning without neurons in physical systems

Learning without neurons in physical systems

Soft matter roadmap*

Circuits that train themselves: decentralized, physics-driven learning

Contact Info

Product

Resources

About

Soft matter roadmap^*