Artificial Intelligence Strategies for Materials Discovery with Explainable & OpenSharing Capabilities by Prof. Prasanna V. Balachandran

In this talk, the Prof discussed some of the past and ongoing research in the application of machine learning methods to efficiently guide materials design and discovery efforts. The overarching theme is efficient navigation of the vast search space, which is especially critical when brute-force evaluation of the search space is prohibitively expensive. Adaptive learning methods, such as active learning and Bayesian optimization, provide a promising solution to address this important problem.

One of the expected outcomes from an iterative adaptive learning loop is an improved black-box machine learning or surrogate model that is believed to capture the complexity of the structure-property relationships with sufficient accuracy. More recently, our group has expanded the adaptive learning paradigm in two directions:

(1) Incorporating novel post-hoc model explainable methods to peek inside the trained
black-box models and explain the predictions for each observation in the data set.

(2) Build Web Applications that will allow the public to interact with our trained models
and accelerate discoveries.

Seminar Link: Picoelectrodynamics Theory Network - YouTube