Mechanical regularization

TL;DR

This paper introduces a regularization-inspired approach to improve the training of materials with complex elastic functions, enhancing robustness, capacity, and convergence speed by constraining the search space.

Contribution

It demonstrates that geometrical constraints act as regularizers, preventing failure modes and enabling training of more complex responses with faster convergence.

Findings

Constraining the search space increases robustness and capacity.

Geometrical constraints prevent spurious low-frequency modes.

Regularization improves convergence and response complexity.

Abstract

Training materials through periodic drive allows to endow materials and structures with complex elastic functions. As a result of the driving, the system explores the high dimensional space of structures, ultimately converging to a structure with the desired response. However, increasing the complexity of the desired response results in ultra-slow convergence and degradation. Here, we show that by constraining the search space we are able to increase robustness, extend the maximal capacity, train responses that previously did not converge, and in some cases to accelerate convergence by many orders of magnitude. We identify the geometrical constraints that prevent the formation of spurious low-frequency modes, which are responsible for failure. We argue that these constraints are analogous to regularization used in machine learning. Our results present a unified understanding of the relations between complexity, degradation, convergence, and robustness.