Correlated Optical Reservoir Computing

Reservoir computing is a simple but efficient recurrent neural network, which has been applied to many kinds of learning task. However, its capability is expected to improve further, considering the rapid growth of data in recent years. By contrast, encoding data into the correlation of particles has turned out to be a great strategy for enhancing the processing of information, a typical example of which is the schemes of quantum reservoir computing. Although the schemes have been investigated much in theories, their experimental implementations keep challenging due to the unsatisfying quality of the quantum gates in the current stage. Here, we propose theoretically and demonstrate experimentally a scheme that uses quantum-inspired classical correlation of light to encode the data of learning tasks, as a new way of reservoir computing. This scheme combines the advantages of correlations for computing and the stability of the classical system for generating the correlations, which is called the correlated optical reservoir computing. Our experimental results are in excellent agreement with theoretical calculations, demonstrating the soundness and feasibility of the scheme. Our proposal opens a new avenue for enhanced reservoir computing, which will benefit information processing in the era of big data.