Abstract
Most materials are changed by their history and show memory of things past. However, it is not clear when a system can continually learn new memories in sequence, without interfering with or entirely overwriting earlier memories. Here, we study the learning of multiple stable states in sequence by an elastic material that undergoes plastic changes as it is held in different configurations. We show that an elastic network with linear or nearly linear springs cannot learn continually without overwriting earlier states for a broad class of plasticity rules. On the other hand, networks of sufficiently nonlinear springs can learn continually, without erasing older states, using even simple plasticity rules. We trace this ability to cusped energy contours caused by strong nonlinearities and thus show that elastic nonlinearities play the role of Bayesian priors used in sparse statistical regression. Our model shows how specific material properties allow continual learning of new functions through deployment of the material itself.
4 More- Received 28 May 2020
- Accepted 29 June 2020
DOI:https://doi.org/10.1103/PhysRevX.10.031044
Published by the American Physical Society under the terms of the Creative Commons Attribution 4.0 International license. Further distribution of this work must maintain attribution to the author(s) and the published article’s title, journal citation, and DOI.
Published by the American Physical Society
Physics Subject Headings (PhySH)
Popular Summary
Mechanical metamaterials—composites of multiple elements arranged to exhibit properties not found in natural materials—are usually designed to show one fixed set of behaviors. These structures typically have to be redesigned from scratch to accommodate each new function. In contrast, biological networks, such as the brain, can grow and acquire new functions over time as needed. Continual acquisition of new functions, without forgetting existing ones, has been studied recently in neural networks, but the mechanical requirements for continual learning are not known. In this paper, we explore a learning framework where elastic networks, as they are used, learn multiple stable states in sequence by growing new bonds.
Inspired by cellular cytoskeleton and microtubules as well as recent synthetic nanotubes, we study an elastic network whose links grow and strengthen over time according to the local geometry of the network. We find that such a network can acquire desired stable states in sequence only if the elastic elements are sufficiently nonlinear. In contrast, a top-down centrally planned design of the same states does not require such nonlinearity but cannot acquire new stable states.
Our work clarifies the kind of local interactions needed for synthetic and natural self-healing materials to learn new functions in an ongoing manner without erasing prior ones. Such continuously adapting structures could develop new functionality through deployment, as required by users.