Title: Optical computing with waveguides and complex media
Abstract: The growing computational demands of modern machine learning call for new physical computing paradigms that go beyond digital electronics. Optical systems offer intrinsic advantages, massive parallelism, high bandwidth, and low latency, making them compelling candidates for next-generation computing hardware. In this talk, I will present our recent work on harnessing spatiotemporal nonlinear dynamics in multimode fibers, integrated waveguides, and other complex photonic media for optical information processing. I will discuss how nonlinear light propagation through these systems can implement key computational primitives, including random projections, recurrent transformations, and spiking dynamics. Specifically, I will cover our demonstrations of scalable optical learning operators, genetically programmable optical random neural networks, fiber-based diffractive deep neural networks, and recurrent neural networks realized through spatiotemporal light propagation. I will also present our efforts on integrated platforms, including nonlinear pulse dynamics in multimode silicon nitride waveguides and engineered rogue waves via multimode interactions in photonic waveguides. Finally, I will discuss self-optimizing multichannel architectures and the role of spatiotemporal chaos in shaping the computational capacity of photonic neural networks. These results collectively outline a path toward energy-efficient, scalable, and programmable optical computing with complex media.
