Unified Photonic Implementation of Reservoir Computing and Extreme Learning Machines based on a Single Time-delayed Node
Silvia Ortín, Daniel San-Martín, Luis Pesquera, Miguel C. Soriano, Daniel Brunner, Ingo Fischer, Cláudio R. Mirasso, José Manuel Gutiérrez · 2015
In recent years two machine learning approaches, Extreme Learning Machines (ELM) [1] and Reservoir Computing, in particular Echo State Networks (ESN) [2], have attracted great interest for information processing because of their simplifying training process. Both approaches are based on random nonlinear projections of data into a high-dimensional network using an intermediate single layer of neurons. In ELM the neurons are not inter-connected, while in ESN connectivity provides the fading memory suitable for time-dependent data.