Search for: All records

Physical reservoir computing leverages the intrinsic history-dependence and nonlinearity of hardware to encode spatiotemporal signals directly at the sensor level, enabling low-latency processing of dynamic inputs. Encoding delity depends on the separability of multi-state outputs, yet in practice it is often hampered by empirically chosen, suboptimal operating conditions. Here, we apply Bayesian optimization to improve the encoding performance of solution-processed Al₂O₃/In₂O₃ thin- lm transistors. By exploring a ve-dimensional pulse-parameter input space and using the normalized degree of separation for output state distinguishability, we demonstrate high- delity 6-bit temporal encoding corresponding to 64 output states. We further show that a model based on simpler 4-bit data can effectively guide optimization for more complex 6-bit tasks, substantially reducing experimental effort. Using a six-frame moving-car image sequence as a benchmark, we nd that the optimized 6-bit pulse conditions signi cantly enhance encoding accuracy, with 4-bit derived parameters performing comparably in terms of pixel errors. Shapley Additive Explanations (SHAP) analysis further reveals that gate-pulse amplitude and drain voltage are the dominant contributors to output state separation. This work establishes a data-driven strategy for identifying optimal operating conditions in reservoir devices and outlines a framework that can be transferred to diverse material platforms and physical reservoir implementations. 

Abstract This article presents a novel approach for generating metamaterial designs by leveraging texture information learned from stochastic microstructure samples with exceptional mechanical properties. This eXplainable Artificial Intelligence (XAI)-based approach reduces the reliance on brainstorming and trial-and-error in inspiration-driven design practices. The key research question is whether the texture information extracted from stochastic microstructure samples can be used to design metamaterials with periodic structural patterns that surpass the original stochastic microstructures in mechanical properties. The proposed approach employs a pretrained supervised neural network and applies the Activation Maximization Texture Synthesis (AMTS) method to extract representative textures from high-performance stochastic microstructure samples. These textures serve as building blocks for creating novel periodic metamaterial designs. Using three benchmark cases of stochastic microstructure-inspired periodic metamaterial design, we compare the proposed approach with an earlier XAI design approach based on Gradient-weighted Regression Activation Mapping (Grad-RAM). Unlike the proposed approach, Grad-RAM extracts local microstructure patches directly from the original sample images rather than synthesizing representative textures to generate novel periodic metamaterial designs. Both XAI-based design approaches are evaluated based on the mechanical properties of the resulting designs. The relative merits of both approaches in terms of design performance and the need for human intervention are discussed.