AI and the World's Hard Problems · Chemistry & materials · arXiv · 2026-10-21
Interpretable Meta-Learning for Multi-Objective Chemical Search
The authors present a modular pipeline that combines interpretable linear meta-learning models with adaptive-confidence uncertainty quantification in an Efficient Global Optimization framework for multi-objective molecular discovery. Meta-learned surrogates trained across chemical objectives and cheap auxiliary properties acquire transferable chemical knowledge that adapts rapidly to new objectives from limited data, and in a real large-scale search for spin-crossover metal-organic complexes the baseline performs 78% worse in the Pareto sense. An adaptive confidence-tuning algorithm that dynamically recalibrates the exploration-exploitation trade-off as the search evolves further dominates over 50% of the statically calibrated front.