Towards Deep Physical Reservoir Computing Through Automatic Task\n Decomposition And Mapping

Photonic reservoir computing is a promising candidate for low-energy\ncomputing at high bandwidths. Despite recent successes, there are bounds to\nwhat one can achieve simply by making photonic reservoirs larger. Therefore, a\nswitch from single-reservoir computing to multi-reservoir and even deep\nphysical reservoir computing is desirable. Given that backpropagation can not\nbe used directly to train multi-reservoir systems in our targeted setting, we\npropose an alternative approach that still uses its power to derive\nintermediate targets. In this work we report our findings on a conducted\nexperiment to evaluate the general feasibility of our approach by training a\nnetwork of 3 Echo State Networks to perform the well-known NARMA-10 task using\ntargets derived through backpropagation. Our results indicate that our proposed\nmethod is well-suited to train multi-reservoir systems in a efficient way.\n