Towards Trainable Media: Using Waves for Neural Network-Style Training
This work proposes a novel approach for building trainable hardware systems, potentially enabling faster and more efficient neural network training in integrated photonics.
The paper tackles the problem of implementing neural network-style training in physical media by using wave interactions to perform matrix-vector multiplication, demonstrating a numerical example where gradient computation is achieved through signal and error waves.
In this paper we study the concept of using the interaction between waves and a trainable medium in order to construct a matrix-vector multiplier. In particular we study such a device in the context of the backpropagation algorithm, which is commonly used for training neural networks. Here, the weights of the connections between neurons are trained by multiplying a `forward' signal with a backwards propagating `error' signal. We show that this concept can be extended to trainable media, where the gradient for the local wave number is given by multiplying signal waves and error waves. We provide a numerical example of such a system with waves traveling freely in a trainable medium, and we discuss a potential way to build such a device in an integrated photonics chip.