Quantum reservoir computing with linear photonic networks

We propose a platform for quantum machine learning based on photon number resolving (PNR) detection and linear optical networks which is implementable with consumer technology. We use a reservoir computing architecture which allows the system to be realized with simple optical components and reduces the computational cost of training while PNR detection enables scaling of the computational power without increasing the physical complexity of the network. We show that beyond PNR, increasing the quantum character of the input state yields improved performance and we introduce novel figures of merit to quantify and interpret this quantum enhancement.