Mechanical Neural Networks

 Technical intro abstract

Mechanical neural networks: Architected materials that learn behaviors

RYAN H. LEE HTTPS://ORCID.ORG/0000-0003-2509-1677, ERWIN A. B. MULDER, AND JONATHAN B. HOPKINS HTTPS://ORCID.ORG/0000-0003-4752-746X Authors Info & Affiliations

SCIENCE ROBOTICS  19 Oct 2022, Vol 7, Issue 71, DOI: 10.1126/scirobotics.abq7278

Abstract

Aside from some living tissues, few materials can autonomously learn to exhibit desired behaviors as a consequence of prolonged exposure to unanticipated ambient loading scenarios. Still fewer materials can continue to exhibit previously learned behaviors in the midst of changing conditions (e.g., rising levels of internal damage, varying fixturing scenarios, and fluctuating external loads) while also acquiring new behaviors best suited for the situation at hand. Here, we describe a class of architected materials, called mechanical neural networks (MNNs), that achieve such learning capabilities by tuning the stiffness of their constituent beams similar to how artificial neural networks (ANNs) tune their weights. An example lattice was fabricated to demonstrate its ability to learn multiple mechanical behaviors simultaneously, and a study was conducted to determine the effect of lattice size, packing configuration, algorithm type, behavior number, and linear-versus-nonlinear stiffness tunability on MNN learning as proposed. Thus, this work lays the foundation for artificial-intelligent (AI) materials that can learn behaviors and properties.  .. ' 

Solving Problems with 2D Materials

Note the mention of 'simulated annealing' methods.   

Solving the 'Big Problems' via Algorithms Enhanced by 2D Materials

Penn State News. Jamie Oberdick, January 5, 2022

Pennsylvania State University (Penn State) researchers have developed a method of solving combinatorial optimization problems using two-dimensional (2D) materials. The researchers utilized a simulated annealing algorithm to determine the ground state of an Ising spin glass system. Penn State's Amritanand Sebastian said the process involves conducting in-hardware computational operations, with the hardware deployed via 2D material-based transistors that also store data. "We make use of this in-memory computation capability in order to perform simulated annealing in an efficient manner," he explained. According to Sebastian, the method saves energy through ultra-low-power operation, allows efficient computation of the spin system's energy, and does not require the hardware to scale with the size of the problem.  ... '