Learning from Conjugated Polymers: Statistical Data Mining from 50 Million and Counting

Geoffrey Hutchison
University of Pittsburgh

Conjugated organic polymers represent an interesting test case for computational materials design, since the search space is vast and property prediction requires some flavor of time-consuming quantum chemical calculation. Moreover, many of the physical scaling laws are known either from first-principals or empirical experimental and computational work. I'll discuss our efforts to use genetic algorithm discrete optimization to rapidly find optimal and near-optimal targets for organic solar cells, statistical data mining efforts to design physically-motivated heuristics for scaling laws, and challenges for machine learning and such statistical approaches, including the molecular conformation problem.

Back to Workshop IV: Synergies between Machine Learning and Physical Models