A Survey on Reservoir Computing and its Interdisciplinary Applications Beyond Traditional Machine Learning

Reservoir computing (RC), first applied to temporal signal processing, is a\nrecurrent neural network in which neurons are randomly connected. Once\ninitialized, the connection strengths remain unchanged. Such a simple structure\nturns RC into a non-linear dynamical system that maps low-dimensional inputs\ninto a high-dimensional space. The model's rich dynamics, linear separability,\nand memory capacity then enable a simple linear readout to generate adequate\nresponses for various applications. RC spans areas far beyond machine learning,\nsince it has been shown that the complex dynamics can be realized in various\nphysical hardware implementations and biological devices. This yields greater\nflexibility and shorter computation time. Moreover, the neuronal responses\ntriggered by the model's dynamics shed light on understanding brain mechanisms\nthat also exploit similar dynamical processes. While the literature on RC is\nvast and fragmented, here we conduct a unified review of RC's recent\ndevelopments from machine learning to physics, biology, and neuroscience. We\nfirst review the early RC models, and then survey the state-of-the-art models\nand their applications. We further introduce studies on modeling the brain's\nmechanisms by RC. Finally, we offer new perspectives on RC development,\nincluding reservoir design, coding frameworks unification, physical RC\nimplementations, and interaction between RC, cognitive neuroscience and\nevolution.\n