Combining machine learning and human understanding for accelerated discovery of molecular materials

Machine learning has been widely used to accelerate materials design; however, despite intensive research, almost all materials are still being discovered using heuristic design rules based on theory and experimentation. This raises a question if fully autonomous materials design is possible at all or whether it is a process that requires intellectual involvement. Or maybe there is a middle ground where both approaches can meet and reinforce each other? This proposal puts forward an idea of using model Hamiltonians combined with machine learning and human understanding to formulate new design rules for functional materials. We propose to expedite the human learning process by using algorithms to find patterns which can be transformed into new design rules. If successful, this approach could be a paradigm shift in accelerated materials discovery. The development of methodology will be driven by a concrete example in organic optoelectronics.