In 2022, Dr. Ulisses Braga-Neto, Director of the TAMIDS SciML Lab, co-authored a paper with his doctoral candidate, Levi McClenny, that is now getting significant recognition in the scientific community. “Self-adaptive physics-informed neural networks” recently surpassed 1,000 citations on Google Scholar!
At the time, Dr. McClenny was an active member of the U.S. Army Reserve and a Black Hawk helicopter pilot based in Conroe, Texas. During his time at Texas A&M, his studies were funded by a National Science Foundation Research Traineeship (NRT) grant under Dr. Raymundo Arroyave, PI and associate professor in the Department of Materials Science and Engineering. Drs. McClenny and Braga-Neto worked on research projects focused on using state-of-the-art machine learning tools to better understand what happens to materials at a microscopic level. Dr. McClenny aimed to apply this knowledge to military vehicles and other structures to help predict and prevent deterioration and breakage.
Their 2022 paper, published in the Journal of Computational Physics, introduced a new way to train a type of AI model known as physics-informed neural networks (PINNs). These models are designed to solve complex physics equations, but they often struggle with the most difficult parts of a problem. This new, innovative method allows the model to adapt during training by assigning greater importance, or weight, to areas where it performs poorly. To ensure a balanced learning process, they applied a statistical method called Gaussian Process regression to spread the PINNs’ “attention” to surrounding points instead of treating each point in isolation.
The result? Their method solved these complex physics problems faster and more accurately than existing approaches. A paper published in 2025 by Dr. George Karniadakis, a pioneer in the field of machine learning, listed self-adaptive PINNS as “one of the key advancements in physics-informed machine learning.”
This milestone not only highlights the impact of Drs. McClenny and Braga-Neto’s work, but also the growing importance of improving intelligent systems. By training these models to focus on the most difficult areas of scientific discovery, self-adaptive PINNs represent a meaningful step forward in solving complex problems accurately, quickly, and efficiently.