Invited and hosted by Professor Yu-Lun Liu, Dr. Peter Yichen Chen, Assistant Professor in the Department of Computer Science at the University of British Columbia (UBC) and Director of the PhysAI Lab, recently delivered a keynote lecture titled "Neural PDE: AI-Enhanced Physics Simulation." In his talk, Dr. Chen guided participants to the emerging frontier at the intersection of artificial intelligence and physics simulation, sharing the latest breakthroughs from his research team in this domain.

Dr. Chen began his lecture with a thought-provoking question: What if we could precisely predict the trajectory of a typhoon, simulate in real time how a new drug interacts within the human body, or design safer and more efficient aircraft? Such possibilities, he emphasized, all depend on powerful physics simulation technologies. Today, physics simulation is widely regarded as the "third pillar" of modern science—standing alongside theoretical derivation and experimental observation—as an indispensable tool for understanding nature and driving applications.

For decades, scientists have relied on two distinct approaches to physics simulation. The first, the "classical" approach, is rooted in the theories of Newton, Einstein, and others, employing mathematical formalisms such as partial differential equations to describe natural phenomena—like a handbook of physics that rigorously explains how the world works. While reliable, this method demands intensive computation and deep mathematical expertise. The second, the "data-driven" approach, has surged in recent years, leveraging neural networks to extract patterns directly from massive observational datasets. Much like how a computer trained on millions of images can recognize a cat, data-driven methods offer speed and flexibility. However, when faced with entirely new or unseen scenarios, they often fail to provide robust predictions.

To overcome these limitations, Dr. Chen proposed a solution: combining physical laws with artificial intelligence to create "hybrid simulation systems." In this framework, physical laws serve as inductive biases that guide AI learning, ensuring that the models do more than memorize data—they also capture underlying principles. This is akin to giving students not just practice tests, but also a textbook that helps them grasp conceptual structures, enabling them to make sound judgments even in unfamiliar contexts.

This hybrid paradigm demonstrates striking advantages. First, it captures fine-grained details that traditional equations struggle to model, making simulations more realistic. Second, it dramatically boosts efficiency—computations that once required days on a supercomputer can now be performed in hours. Third, it lowers the barrier to entry, empowering researchers without specialized backgrounds in physics or mathematics to leverage these tools effectively. Such progress not only opens new opportunities for fundamental science but also accelerates cross-disciplinary applications.

Dr. Chen emphasized that hybrid simulation is more than just a technical advance—it may fundamentally transform the way scientific discovery is pursued. From climate change studies and drug design to human modeling, new materials research, and aerospace innovation, this approach holds vast potential. It enables unprecedented speed and accuracy in understanding nature, while also producing tangible real-world impact.

Currently an Assistant Professor at UBC and Director of the PhysAI Lab, Dr. Chen previously conducted postdoctoral research at MIT CSAIL. He earned his Ph.D. in Computer Science from Columbia University and earlier studied mathematics at UCLA, where he received the Sherwood Prize. His research spans computer graphics, machine learning, scientific computing, mechanics, and robotics, with a focus on advancing 3D content creation, engineering design and control, and materials discovery. His contributions have received international recognition, including the Best Paper Award at SIGGRAPH.

This special lecture, hosted by Professor Yu-Lun Liu, not only showcased the latest advances in AI-enhanced physics simulation but also highlighted the importance of interdisciplinary collaboration. As this line of research continues to develop, both academia and industry are stepping into a new era of faster, more accurate, and more accessible simulation—realizing the vision Dr. Chen outlined: hybrid physics–data simulation systems opening a new chapter in scientific exploration.