Optical Generative Models & In Situ Learning
Generative AI and reinforcement learning implemented in the optical domain
Bringing modern machine learning paradigms into the optical domain: optical generative models (Nature, 2025) demonstrate snapshot image generation through light diffraction, inspired by diffusion models; and model-free optical processors (Light: Science & Applications, 2026) are trained in situ with reinforcement learning (proximal policy optimization), requiring no digital twin of the physical system.
I also contributed to the broader effort on the training of physical neural networks (Nature, 2025) and spectral kernel machines with electrically tunable photodetectors (Science, 2025).