Guest Seminar by Department of Chemical and Biological Engineering - Accelerating Material and Chemical Discovery with Deep Neural Networks and Active Learning

The advancement of material and chemical development is critical to global sustainable development. Due to its potential to revolutionize this process, the idea of using data science in chemical engineering to accelerate advanced material and chemical discovery has gained vast interests. Despite the enormous potential demonstrated by data science, its application in real advanced material and chemical discovery processes is usually hampered by data scarcity. In this seminar, we will provide an overview of our research efforts to develop domain-inspired active learning, data augmentation, and deep learning system design to resolve material and chemical discovery challenges at various scales.

First, we will discuss how expert-guided molecular space generation algorithm and target property focused active learning can help with micro-scale functional material property optimization. Second, an expert informed three-stage ML framework will be discussed to address the challenge of customized complex device design. Finally, a data-efficient molecule image processing framework will be introduced to automate the characterization process. These techniques for material/chemical property and synthesis recipe prediction, efficient data collection, and instrument automation are integrated into an overall workflow that identifies a tractable data collection space for a given material and chemical discovery challenge, conducts intelligently guided data collection, automates the collection process if possible, and optimizes the final model for the design target.