The first step that project aim is to deliver a new method for providing the most suitable nonlinear response of the network node. Based on this nonlinearity we will propose several novel and advanced architectures of all-optical reservoir computer. The second step is to experimentally verify our theoretical findings regarding proposed nonlinearity. Finally, we will use optical computer and benchmark it, in a case-study of financial market time-series prediction.

The nonlinearity proposed in our project will belong to most desirable and favorable class in the field of neural networks. However, its most exciting and attractive feature is the adaptivity. The nonlinearity can change the shape from various sigmoid to various PReLU types by slightly changing operational parameters. This means that a single component can be used not only in RC but also in other photonic neural networks. The multiple input in the nonlinear node, might lead to softmax activation function, suitable for transformer networks.

The novel design of nonlinear node, will open playground for more advanced design of all optical RC. We will have opportunity to investigate different network configuration based on coupled nonlinear nodes, or multiple injection providing increase of parallelization, speed increase and performance quality improvement.