A Machine Learning Workflow for Material Design Optimization

Designing materials requires experimentation with many different potential designs. Even with computational simulations as aids, the computational cost of running many simulations limits the ability to explore the design space. A machine-learning-based workflow is developed for material design that can be adapted to a variety of optimization problems. By training a deep neural network, simulation results can be rapidly approximated, allowing for more efficient searches of the parameter space and faster convergence to an optimal solution. We show that this workflow can efficiently find optimized material designs while reducing the number of expensive simulations.