Title: Unsupervised Machine Learning for Application-specific Materials Search.
Presenter: Dr. Shehrin Sayed (UC Berkeley)
Abstract: Every year, many exciting scientific results get published in various journals from different disciplines. A particular material or a materials combination is often discussed in different contexts, and non-trivial connections among scientific results can get hidden within the overwhelmingly large published text. Recently, there has been a growing interest in natural language processing algorithms to bring the enormous scientific literature under the same umbrella.
In this talk, I'll discuss a new approach using an unsupervised machine learning model that has been trained using unlabeled text collected from published scientific abstracts in materials science, physics, and engineering journals. The trained model encodes information into word embeddings or high-dimensional vectors, and the relationships among these embeddings reflect the knowledge on various materials properties, related phenomena, and applications. More interestingly, the model identified the intrinsic figure-of-merit of various materials for a particular phenomenon or application, just from the correlations of one word to the other words within the corpus used for training. The patterns identified from such correlations reasonably match the patterns obtained from experimental results. In addition, the model identified several new materials associated with a given phenomenon, not yet been studied in the field. Although the model applies to diverse topics, I'll show some examples of emerging magnetic and spintronic materials and phenomena.
Bio: Shehrin Sayed is a Postdoctoral Researcher in Professor Sayeef Salahuddin's group and is jointly affiliated with the Department of Electrical Engineering and Computer Sciences at the University of California-Berkeley and the Materials Sciences Division in Lawrence Berkeley National Laboratory. He received his Ph.D. in 2018 and M.Sc. in 2013 from the Department of Electrical and Computer Engineering at Purdue University. His Ph.D. dissertation received the 2018 award by Dimitris N. Chorafas foundation, characterized by its potential for practical application. His research focuses on developing efficient transport models and simulation frameworks that bridge emerging materials to novel device structures and analyze their performance within conventional circuits. His interest includes magnetic, spintronic, 2D, and quantum materials for significant improvements in nonvolatile memory and energy harvesting applications.
ASCENT task: 2776.021: Current Driven (Anti)Ferro-Magnetic Memory
This meeting is only available to the JUMP research community, such as Principal Investigators, Postdoc researchers, Students, and Industry/Government liaisons.