| Abstract: | Learning is traditionally studied in biological or computational systems. The power of learning frameworks in solving hard inverse-problems encourages the development of `physical learning machines' - physical systems that adopt desirable properties on their own, without computational design. It was recently discovered 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 present recent work in the emerging field of physical learning, describing theoretical and experimental advances in systems ranging from flow and electrical networks to mechanical materials. Such physical learning machines can learn to solve problems inspired by biology, material science and computational machine learning. These machines provide multiple practical advantages over computer designed systems, including not requiring an accurate model of the system, robustness to damage, and the ability to adapt to changing needs over time. As theoretical constructs, physical learning machines afford a novel perspective on how physical constraints modify abstract learning theory. |
