Seminar by Professor Hangwen GUO (Fudan University)

With the ever-growing demand for AI computing power, conventional digital logic architectures face fundamental bottlenecks of energy efficiency, making it urgent to explore entirely new computing paradigms. Physical systems, with their intrinsic parallelism, nonlinearity, tunable fluctuations, and ultra-low power consumption, offer tremendous advantages and are giving rise to the emerging field of “Physics for AI.” This concept harnesses physical principles and dynamical processes to perform computation and solve problems directly in hardware, bypassing parts of conventional algorithmic processing. In this talk, I will approach the topic from the perspective of magnetism, complemented by elements of statistical physics, to realize a self-learning Hopfield neural network and address optimization problems using Ising machines. In addition, I will share a few examples of using physical systems to achieve ultrafast dynamic computing for image processing and backpropagation-free training of neural networks, highlighting their unique advantages and potential applications.

You may also like the following events from Department of Physics, CUHK 香港中文大學物理系:

• Next month, 5th September, 04:00 pm, Colloquium by Dr. Howard HUI (Caltech) in Hong Kong

Also check out other Workshops in Hong Kong, Photography events in Hong Kong.