Benchmarking the Acceleration of Materials Discovery by Sequential Learning

Abstract

Sequential learning (SL) strategies, i.e. iteratively updating a ma-chine learning model to guide experiments, have been proposed to significantly accelerate materials discovery and research. Applications on computational datasets and a handful of optimization experiments have demonstrated the promise of SL, motivating a quantitative evaluation of its ability to accelerate materials discovery, specifically in the case of physical experiments. The benchmarking effort in the present work quantifies the performance of SL algorithms with respect to a breadth of research goals: discovery of any "good" material, discovery of all "good" materials, and discovery of a model that accurately predicts the performance of new materials.

Two-Step Machine Learning Enables Optimized Nanoparticle Synthesis

“…However, most prior work focuses on either optimization which lacks interpretability and transferability when the target changes 5 or inverse design using regression which uses a static dataset 20 . Algorithm selection using information criteria such as Akaike information criterion (AIC) 26 and Bayesian information criterion (BIC) 27 could be used to maximize time-and resource-efficiency of closed-loop laboratories, e.g., by leveraging co-evolution, physicsfusion, and related strategies [28][29] . While it might not be the only possible ML architecture for such a problem, our approach attempts to combine efficient optimization, interpretability and knowledge extraction.…”

Benchmarking the Performance of Bayesian Optimization across Multiple Experimental Materials Science Domains

“…To quantify the acceleration of discovery from BO, we adapt two other metrics similar to the ones from Rohr et al [28].…”

“…is the ratio of cycle numbers showing how much faster one could reach a specific value Top%(i BO ) = Top%(i random ) = a ∈ [0, 1]. The aggregated performance of BO algorithms is further quantified via EF and AF curves in Figure 3 algorithm selection varies with assigned experiment budget and specific optimization task [28].…”

“…There have been very few quantitative analyses of the acceleration or enhancement resulting from applying BO algorithms and discussions on sensitivity of BO performance to surrogate model and acquisition function selection. Rohr et al [28], Graff et al [29] and Gongora et al [24] have evaluated the performance of BO using multiple surrogate models and acquisition functions within specific electrocatalyst, ligand and mechanical structure design spaces respectively. However, comprehensive benchmarking of the performance of BO algorithms across a broad array of experimental materials systems, as we present here, has not been done.…”

Report of the Basic Energy Sciences Roundtable on Liquid Solar Fuels

“…Theory-guided, high-throughput experimental discovery has recently been applied successfully to complex materials discovery problems in both photoelectrochemistry 183 and flow batteries 184 . The large data sets that are emerging from these studies now offer opportunities for machine learning and prediction of new material combinations with enhanced properties 185 .…”