Active Learning and Explainable AI for Multi-Objective Optimization of Spin Coated Polymers

Active learning reduces experimental waste: The framework uses PyePAL to intelligently select which polymer samples to test next, focusing on the most promising regions of the design space rather than testing randomly. This adaptive approach means researchers need fewer experiments to find good solutions—a significant practical benefit when experiments are time-consuming or expensive.

AI explanations help experts make decisions: Rather than treating the optimization as a "black box," the researchers use fuzzy linguistic summaries to translate complex relationships between processing parameters (spin speed, dilution, polymer mixture) and outcomes (hardness and elasticity) into human-readable statements. This allows materials scientists to understand why certain designs work, not just that they work.

Visualization makes trade-offs transparent: The use of UMAP to visualize the Pareto front in two dimensions helps stakeholders see the fundamental trade-offs in the problem—for instance, that improving hardness might require sacrificing elasticity. This transparency is crucial for responsible decision-making when multiple objectives compete.

Multi-objective optimization reflects real-world complexity: Rather than optimizing for a single "best" outcome, this framework acknowledges that polymer coatings need to balance multiple properties simultaneously. This realistic framing helps prevent unintended consequences that can arise when optimization focuses too narrowly on one metric.