Speaker
Paula Turkiewicz
(Adam Mickiewicz University)
Description
Neuromorphic computing aims to perform machine-learning tasks directly in physical hardware, exploiting the system's own dynamics instead of digital simulation. Wanjura and Marquardt recently proposed realizing neural networks in a physical system based on linear wave scattering, showing that a nonlinear activation function can emerge from purely linear optics [1]. Here, we test this idea by implementing a simple neural network, based on linear wave scattering, that performs a binary classification task: the XOR logic operation, a classic example of a non-linearly separable problem.
[1] C. C. Wanjura, F. Marquardt, Nat. Phys. 20, 1434–1440 (2024).
Author
Paula Turkiewicz
(Adam Mickiewicz University)