Multifunctional Photonic Reservoir Computing Based on Semiconductor Laser With Optical Feedback

The human brain can efficiently handle multiple tasks simultaneously, with the structural segregation of brain regions closely linked to their functional specialization. In this paper, we propose and experimentally demonstrate multifunctional photonic reservoir computing (MPRC), inspired by this biological characteristic. The neuron states of the photonic reservoir are divided into distinct sections for different tasks by the designing of input matrix, and only a set of output weight matrix, applicable to all tasks, is ultimately trained. Therefore, MPRC is capable of handling multiple tasks simultaneously with a fixed set of parameters, mitigating the tedious process of hyperparameter optimization. The influence of neuron partition size and operating parameters on MPRC performance is studied systematically. Furthermore, the generalizability of MPRC is validated by performing four different tasks simultaneously. Finally, our results are experimentally demonstrated using semiconductor lasers with a time-delay feedback loop.