Presentation Information
[8p-N205-4]Sub-quadratic scalable approximate matrix multiplier
using multi-plane light conversion
〇Yoshitaka Taguchi1 (1.Tokyo Univ.)
Keywords:
Optical computing,Optical matrix-vector multiplication,multi-plane light conversion
Optical matrix–vector multiplication holds great promise for machine learning applications. Programmable computing devices require as many modulators as the degrees of freedom of the matrix, leading to a number of modulators that scales as the square of the number of modes. In this work, we show that by using multi-plane light conversion (MPLC) to implement approximate rather than exact matrix operations, one can construct a matrix-multiplication accelerator whose scaling is sub-quadratic. We validate this approach through numerical simulations on a text classification model.