Challenges of Using Machine Learning/Artificial Intelligence for Real-world Experiments

Prof. Julia W. P. Hsu from Department of Materials Science and Engineering Erik Jonsson School of Engineering and Computer Science Texas Instruments Distinguished Chair in Nanoelectronics University of Texas at Dallas, United States

@ Room 212, PHYSICS/CCMS Building

Abstract:

The release of ChatGPT in November 2022 marked a watershed event for artificial intelligence/machine learning (AI/ML), shaking up scientific research, education, and public perception of AI technologies. Around this time, my group began exploring how Bayesian Optimization (BO) can accelerate identifying optimal synthesis or processing conditions to make materials with desired properties. BO is particularly useful in cases with high-dimensional inputs, costly experiments, or complex relationships between variables. Our work focuses on applying BO for fabricating high-performance solar cells, transparent conducting electrodes, and flexible metal oxide capacitors. While mathematicians and computer scientists often test BO algorithms on synthetic data sets, applying these techniques to real-world experimental data requires dealing with random fluctuation, time-dependent variation, and tool limitations. In this talk, I will discuss examples of how we have adapted computational methods to handle these issues.

<< Back