Large-Scale Optical Reservoir Computing for Spatiotemporal Chaotic Systems Prediction

Scalable optical learning operator

“…A large database of X-ray images of lungs with various diseases including COVID-19 was used to train the single-layer network that classifies the representation of the lungs produced at the output camera. The notion of combining a complex, fixed mapping with a simpler programmable processor to realize a powerful overall system-including the optical implementation of such machines-has been used in support vector machines 9,10 , reservoir computing [11][12][13][14][15] , random mappings [16][17][18][19] and extreme learning machines 20,21 . The nonlinear mapping performed by the MMF is not the same as in any of the earlier approaches.…”

Simulating self-learning in photorefractive optical reservoir computers

“…Within the proposed setup, the photorefractive crystal thus acts as a diffractive element inside the cavity, which introduces the random mixing necessary for a reservoir to function. However, even though this setup resembles typical diffractive optical reservoirs [24][25][26][27] , it is important to note that the proposed setup does not contain any nonlinear elements and hence follows the passive photonic reservoir setup as introduced in Eq. (1).…”

Opportunities in Quantum Reservoir Computing and Extreme Learning Machines

“…It is worth mentioning that the generalization of this framework to a vectorial input, , or output, , is straightforward. [ 28,66 ]…”

“…Despite the simplicity in the training, these methods have been successful in numerous practical applications. [ 24–28 ] Although RC and ELM are both random mapping architectures, [ 23 ] a main difference resides in the fact that RC exploits the natural dynamics of the substrate as an internal memory of past input information while ELM does not. Both RC and ELM are amenable to dedicated hardware implementations in, for example, digital electronics [ 29,30 ] or nonlinear analog systems.…”