Towards interpretable artificial intelligence and first principles computations for designing nanoscale materials and interfaces

Abstract: Accurate modeling of complex nanoscale phenomena requires both an accurate structure representation of the system and a relevant theoretical description. By automating this process and developing robust computational tools, we can accelerate our materials understanding for a wide range of applications. Over the past two decades, the advancements in computing power and artificial intelligence (AI)/machine learning methods enabled us to study large scale systems at an accuracy comparable to density functional theory.

In this talk, I will discuss modeling of specific complex systems such as transition metal oxides and 2D materials, along with different computational tools geared towards novel materials discovery and structure inversion to bridge the gap between theory and experiments. I will show application of these tools to understand structure and properties from STM, STEM and PDF data in collaboration with different experimental groups. By applying these techniques and leveraging machine learning interatomic potentials, we can investigate interfaces in bulk materials complementing experiments, which are important for energy storage and electronic applications. Finally, I will discuss the role of interpretability and reliability in AI models and how we can gain insights from their latent representations.

Bio: Venkata Surya Chaitanya Kolluru is a postdoc at the Center for Nanoscale Materials. He obtained his Ph.D. in Materials Science and Engineering from University of Florida.