Reservoir Computing for Scientific Modeling and Data Analysis

A survey of reservoir computing (RC) as a dynamical framework for learning, classification, prediction, and data analysis is presented. Three reservoir architectures are considered: the echo-state network, a FitzHugh-Nagumo excitable network, and a transformer-inspired reservoir, which are implemented to perform tasks of increasing complexity: learning logical functions, physics-informed solving of the damped harmonic oscillator (including autonomous rollout), image classification with an untrained convolutional front-end, and multi-input RC for CLIP-style language-image pre-training.

Contatto:
gabriele.ciaramella@polimi.it