Colloquium by Sylvain Gigan
Tuesday, January 27, 2026, 16:30
HS C
Students and Academic Staff
We invite you to our colloquium with guest Sylvain Gigan, Sorbonne Université, France; Laboratoire Kastler–Brossel, CNRS–ENS–Collège de France.
Computing with Complexity: Large-Scale Photonic Machine Learning leveraging Disorder
Photonics promises a paradigm shift in information processing, offering ultrafast speeds and low energy consumption. While optical computing has seen a resurgence of interest for machine learning applications, current implementations face a scalability bottleneck: most proof-of-concept devices are restricted to low dimensionalities and shallow, simple architectures.
In this colloquium, I will discuss how we can overcome these limitations by exploiting the physical complexity of light scattering itself. I will demonstrate that multiple scattering in disordered media—usually considered a hindrance to imaging—can be harnessed to perform a critical computational operation: large-scale random matrix multiplication. By mapping input information onto the complex optical transmission matrix of a scattering medium, we can process high-dimensional data at the speed of light and at negligible energy cost. I will present experimental results across various machine learning tasks and discuss the roadmap for extending this concept from single-layer operations to deep, recurrent optical neural networks.
Organizers: K. Erath-Dulitz, H.-C. Nägerl, F. Marleau