Fig. 5: Schematic of offline learning.
From: Neuromorphic overparameterisation and few-shot learning in multilayer physical neural networks
Input data is applied to the physical system producing a set of o(t) features. Feature selection reduces the size of o(t) to a new set of x(t) features. We then apply ridge regression to obtain a set of weights Wo to obtain the transformation/prediction y(t) = Wox(t).
