Evolution-guided Bayesian optimization for constrained multi-objective optimization in self-driving labs

The development of automated high-throughput experimental platforms has enabled fast sampling of high-dimensional decision spaces. To reach target properties efficiently, these platforms are increasingly paired with intelligent experimental design. When solving optimization problems, Bayesian-based optimizers are often chosen for their ability to solve a global optimum while allowing both exploration and exploitation. However, for problems with complex constraints that require achieving multiple conflicting objectives rapidly, Bayesian-based optimizers are less effective, and this necessitates the development of new algorithms. Here, we devise a hybrid Evolution-Guided Bayesian Optimization (EGBO) algorithm that introduces selection pressure to decrease sampling wastage; this not only solves the Pareto Front (PF) efficiently but also achieves better coverage of the PF while limiting sampling in the unfeasible space. The algorithm was developed together with a custom self-driving lab for seed-mediated silver nanoparticle synthesis, targeting 3 objectives (1) optical properties, (2) fast reaction, and (3) minimal seed usage alongside complex constraints. In a parallel experimental campaign in our self-driving lab, we evaluate the performance of EGBO against the q-Noisy Expected Hypervolume Improvement optimizer proposed by Daulton et al. and show 14% better performance with 15 runs of experimentation and 72 datapoints. We also demonstrate EGBO’s good coverage of the PF as well as a better ability to propose feasible solutions. EGBO thus offers an efficient solution for solving constrained multi-objective problems in high-throughput experimentation platforms.