Performance optimization of a reservoir computing system based on a solitary semiconductor laser under...

Area of Science:

• Optoelectronics
• Computational Neuroscience
• Nonlinear Dynamics

Background:

• Reservoir computing (RC) systems offer a powerful framework for processing complex temporal data.
• Semiconductor lasers provide a compact and efficient platform for implementing RC systems.
• A key limitation of simple RC systems is their finite memory capacity (MC).

Purpose of the Study:

• To propose and investigate a simple reservoir computing system based on a solitary semiconductor laser.
• To enhance the memory capacity (MC) of the RC system using auxiliary input and output methods.
• To evaluate the optimized RC system's performance on complex computational tasks.

Main Methods:

• A reservoir computing system was designed using a solitary semiconductor laser with electrical message injection.
• Two auxiliary methods, M-input (weighted sum of past inputs) and M-output (output layer optimization), were introduced.
• The system's performance was numerically investigated, focusing on memory capacity enhancement and task-specific optimization.

Main Results:

• The auxiliary methods, particularly the combined M-input and M-output (M-both), significantly improved the system's memory capacity.
• The M-input method proved most effective for the Santa Fe time series prediction task.
• The M-both method demonstrated superior performance for the nonlinear channel equalization (NCE) task.

Conclusions:

• Auxiliary methods effectively enhance the memory capacity of semiconductor laser-based reservoir computing systems.
• The choice of auxiliary method should be tailored to the specific computational task for optimal performance.
• This research paves the way for more robust and efficient optical reservoir computing solutions.