Near-Equilibrium Propagation training in nonlinear wave systems
The episode discusses the paper "Near-Equilibrium Propagation training in nonlinear wave systems," which enables training physical wave systems like polariton condensates. It uses a method that extracts gradients by comparing steady states after a small perturbation (nudging). The approach achieves high accuracy on tasks like MNIST while being vastly faster and more energy efficient than digital simulation, paving the way for physical neuromorphic computing.