Presented By: Chemical Engineering
ChE SEMINAR: Heather Kulik, MIT
"Leveraging experimental data in machine learning and screening to get from computational model to real world materials fast"

A reception with light refreshments will be held in the B10 lobby before each seminar from 1-1:30 p.m.
Abstract: Machine learning in transition metal catalysis is more challenging than other areas of chemistry due to the combination of diversity of chemical bonding and limitations in high quality data sets, experimental or computational. I will describe our efforts to overcome these limitations to accelerate the discovery of novel transition metal containing materials using machine learning. I will discuss how we have leveraged experimental data sets through both text mining and semantic embedding to uncover relationships between structure and function, disseminating high quality datasets of transition metal complexes with known function. I will describe how we've used these data sets to build machine learning models that predict the structure of transition metal complexes. Then I will describe how we have leveraged large datasets of synthesized materials to uncover those with novel function in polymer networks. I will demonstrate the success of our design strategy through macroscopically visible changes in network scale properties of polymers once our transition metal complexes are incorporated. Finally, I will conclude with some outstanding challenges for the field.
Abstract: Machine learning in transition metal catalysis is more challenging than other areas of chemistry due to the combination of diversity of chemical bonding and limitations in high quality data sets, experimental or computational. I will describe our efforts to overcome these limitations to accelerate the discovery of novel transition metal containing materials using machine learning. I will discuss how we have leveraged experimental data sets through both text mining and semantic embedding to uncover relationships between structure and function, disseminating high quality datasets of transition metal complexes with known function. I will describe how we've used these data sets to build machine learning models that predict the structure of transition metal complexes. Then I will describe how we have leveraged large datasets of synthesized materials to uncover those with novel function in polymer networks. I will demonstrate the success of our design strategy through macroscopically visible changes in network scale properties of polymers once our transition metal complexes are incorporated. Finally, I will conclude with some outstanding challenges for the field.