Photonic computation with frequency encoding
Tuesday, January 9, 2024
11 a.m. – 11:45 a.m. ET
This event has been cancelled.
Lingling Fan, Postdoctoral Associate
Electrical Engineering & Computer Science, MIT
Multi-dimensional convolution lies at the cornerstone of artificial intelligence and represents a computationally-intensive step in convolutional neural networks. However, the hardware performance using digital electronics for such convolution operations is constrained by low-speed operation, high-power consumption, and poor scalability to large data.
In her work, Fan theoretically introduces and experimentally demonstrates a multi-dimensional convolution processor via a simple, low-loss setup consisting of a dynamically-modulated ring resonator that is capable of generating arbitrary convolution kernels for frequency-encoded information in the synthetic frequency dimension. Compared with previous optical computational processors based on waveguides, this processor is more compact as part of the required spatial degrees of freedom is now compactly encoded in the modulation tones rather than the spatial coupling constants.
Fan demonstrates that this convolution processor is capable of 2D image convolution and video action recognition. This work points to using compact and reconfigurable integrated photonic circuits to improve machine learning hardware for state-of-the-art artificial intelligence performance.
Attendees can join and participate in the series via Zoom.