Scaling up Echo-State Networks with multiple light scattering

Echo-State Networks and Reservoir Computing have been studied for more than adecade. They provide a simpler yet powerful alternative to Recurrent NeuralNetworks, every internal weight is fixed and only the last linear layer istrained.They involve many multiplications by dense random matrices. Very largenetworks are difficult to obtain, as the complexity scales quadratically bothin time and memory. Here, we present a novel optical implementation ofEcho-State Networks using light-scattering media and a Digital MicromirrorDevice. As a proof of concept, binary networks have been successfully trainedto predict the chaotic Mackey-Glass time series. This new method is fast, powerefficient and easily scalable to very large networks.

放大具有多个光散射的Echo-State网络

回声状态网络和储层计算已经研究了十多年。它们为递归神经网络提供了一种更简单但功能强大的替代方法，每个内部权重都是固定的，并且仅训练最后一个线性层。.. 它们涉及稠密随机矩阵的许多乘法。大型网络很难获得，因为复杂性在时间和内存上都成倍增加。在这里，我们介绍一种使用光散射介质和数字微镜设备的Echo-State网络的新型光学实现。作为概念的证明，已经成功地训练了二进制网络来预测混沌的Mackey-Glass时间序列。这种新方法快速，省电且易于扩展到非常大的网络。 （阅读更多）