Explore generative machine learning approaches for materials discovery in this seminar by Prof. Taylor D. Sparks from the University of Utah. Delve into the potential of machine learning to uncover novel materials that differ chemically and structurally from known examples. Examine generative models like variational autoencoders, generative adversarial networks, and diffusion models, comparing their applications in materials science to image generation. Investigate the unique challenges of generating periodic crystalline structures using these tools. Learn about the Descending from Stochastic Clustering Variance Regression (DiSCoVeR) algorithm, designed to guide the discovery process towards promising yet unintuitive material candidates. Gain insights into the future of materials discovery, including discussions on GAN architecture, encoding distances, uncertainty, and potential next steps in the field.
Moving Beyond Screening via Generative Machine Learning Models for Materials Discovery
Cambridge Materials via YouTube
Overview
Syllabus
Introduction
generative adversarial network
generative models
Gan architecture
Extracting information
Error messages
Sigmoid function
Encoding distances
Layers
Backtracking
Uncertainty
Problems
Where are we
How are we doing
Next steps
Future hope
Taught by
Cambridge Materials
