Associate professor Zhangyang “Atlas” Wang and ph.D. student Peihao Wang have received a DARPA Disruptive Idea Paper Award at the International Conference on Neuro-symbolic Systems (NeuS), 2025. The research team received the award for their paper titled "Why Neural Network Can Discover Symbolic Structures with Gradient-based Training: Theoretic Foundation for Neurosymbolic AI."

NeuS aims to bring together novel concepts, theories, and practices that can help in the development of the science and application of neuro-symbolic computing and systems. The purpose of the DARPA Disruptive Idea Award is to encourage submissions that have the potential to radically change conventional wisdom around AI."

Their new study reveals how neural networks can naturally develop symbolic reasoning abilities—like logic and algebra—through standard gradient-based training. By analyzing how model parameters evolve over time, the researchers show that networks tend to simplify and organize themselves around low-dimensional structures, enabling them to mimic discrete, rule-based thinking. This discovery bridges the gap between neural learning and symbolic reasoning, offering a rigorous theoretical foundation for building more robust, reasoning-capable AI systems.

Zhangyang “Atlas” Wang is a tenured associate professor and holds the Temple Foundation Endowed Faculty Fellowship #7 in the Chandra Family Department of Electrical and Computer Engineering at The University of Texas at Austin. He is also a faculty member of UT Computer Science, and the Oden Institute CSEM program. 

Peihao Wang is doing scientific research in the areas of machine learning and computer vision, under the supervision of Prof. Atlas Wang at the VITA Group. His current research focuses on theories and applications of geometric learning, language models, generative models, and 3D computer vision. He received his bachelor's degree from Shanghai Tech University in 2021.