Reconfigurable and adaptive optical computing

Optical computing offers unparalleled speed and parallelism for linear operations, but its adoption in machine learning has been hindered by the challenge of implementing effective nonlinearities and reconfigurability. This talk presents our latest work in creating optical systems that are not merely programmable, but truly reconfigurable and adaptive. By dynamically controlling light propagation within complex media, such as multimode fibers, diffusers and random matrices, we build computing architectures that can alter their function in real-time to suit different tasks. This approach enables a new class of powerful, energy-efficient photonic processors capable of tackling complex machine learning problems, paving the way for next-generation AI hardware.