Scaling up Echo-State Networks with multiple light scattering
This work addresses scalability issues in Echo-State Networks for researchers in reservoir computing, offering a fast, power-efficient, and easily scalable method, though it is incremental as it builds on existing optical computing approaches.
The authors tackled the problem of scaling Echo-State Networks, which have quadratic complexity in time and memory, by developing an optical implementation using light-scattering media and a Digital Micromirror Device, successfully training binary networks to predict the chaotic Mackey-Glass time series as a proof of concept.
Echo-State Networks and Reservoir Computing have been studied for more than a decade. They provide a simpler yet powerful alternative to Recurrent Neural Networks, every internal weight is fixed and only the last linear layer is trained. They involve many multiplications by dense random matrices. Very large networks are difficult to obtain, as the complexity scales quadratically both in time and memory. Here, we present a novel optical implementation of Echo-State Networks using light-scattering media and a Digital Micromirror Device. As a proof of concept, binary networks have been successfully trained to predict the chaotic Mackey-Glass time series. This new method is fast, power efficient and easily scalable to very large networks.