Data-Driven Optimization Strategies for Accelerating Materials Discovery

In recent years, the integration of high-throughput experimentation with machine learning has revolutionized materials discovery. Here, we present a multitude of case studies in using optimization algorithms for data-driven experiment planning.<br/><br/>We showcase our proposed algorithm Evolution-Guided Bayesian Optimization (EGBO) which integrates a one-step evolution process to mediate exploration and exploitation (J Mat Int, 2023). EGBO shows superior performance in Pareto Front coverage as well as constraint handling, demonstrated for an automated nanoparticle synthesis platform (Npj Comp Mat, 2024).<br/><br/>We also present on challenges and approaches for domain-specific problems. Firstly, preferencing objectives in a multi-objective problem to achieve accurate viscous liquid transfer with minimal transfer times (Digit Discov, 2023). Secondly, efficiently dealing with input constraints in a terpolymer synthesis problem. Lastly, batch-constrained high-throughput sampling for optimization of microring laser fabrication.<br/><br/>Finally, we will also share about successes in implementing both traditional Bayesian optimization as well as multi-task transfer learning for different perovskite synthesis projects (Adv Mat, 2024).