Description:

My new lab at Yale works on two complementary research directions. In the first, we are trying to understand and control photonic systems (i.e., experiments and devices based on light) that are highly multimode. By "highly multimode" I mean that these systems are complex - they have many degrees of freedom, and their physics usually plays out in very high-dimensional spaces. To make the problem even more challenging, we are concerned with the important scenario where all these optical degrees of freedom are interacting, due to nonlinear or quantum effects. Systems of this kind are a wonderful playground for high-dimensional nonlinear physics, but are also practically very important as the foundation for future room-temperature, scalable quantum computers, and for a new generation of advanced light sources (think "Laser 2.0"). In the second direction, we are interested in the general problem of programming complex physical systems. In other words, given some complex physical system with many controllable parameters, how can we "program" it so it does functions we'd like it to do? What we have found is that a deep learning inspired approach - treating complex physical systems almost as if they were artificial neural networks - is well-suited to this. I will summarize our recent work on programming physical systems to create extremely energy efficient analog computers for neural network calculations, and for computational sensing. 

Login Wright, Assistant Professor, Yale University

Host: Istvan Kovacs