LAUSR

Abstract In this communication, we proposed Bayesian optimization acceleration strategy by allowing machine-learned knowledge to flow across different reaction scales. Dispersion polymerization was conducted to validate this knowledge sharing approach. By learning a product uniformity landscape and performing virtual optimization using high-throughput small-scale reaction data, the most promising recipe subspace was extracted from the full-factorial parameter space for batch-size synthesis optimization. The subsequent Bayesian optimizer was hereby able to traverse the subspace and locate a recipe giving large uniform particles (>10 µm). This highly flexible and efficient strategy can be extended to a diverse library of reactions beyond microsphere synthesis.