The physical effects of learning

From electrically responsive neuronal networks to immune repertoires, biological systems can learn to perform complex tasks. In our work, we explore physical learning, a framework inspired by computational learning theory and biological systems, where networks physically adapt to input signals to adopt desired functions without the use of a processor. Unlike traditional engineering approaches, physical learning requires local learning rules. In earlier work, we established a framework for deriving local learning rules for a broad class of physical networks, including some inspired by biological systems, and demonstrated that these rules are physically realizable through laboratory experiments. Here I will describe how learning induces architectural changes in the physical network and a remodeling of its energy landscape, leading to a decrease in the effective physical dimension and a realignment of low eigenvectors of the energy Hessian with the learned task. These effects suggest a method for discovering the task that a novel network may have been trained for.