Enhanced performance of a reservoir computing system based on a dual-loop optoelectronic oscillator

Area of Science:

• Optoelectronics
• Computational Neuroscience
• Machine Learning

Background:

• Time-delayed reservoir computing (RC) offers a brain-inspired approach for temporal information processing.
• Traditional RC systems often require complex architectures, limiting practical implementation.

Purpose of the Study:

• To propose and evaluate a novel, compact, and easily implementable RC system using a dual-loop optoelectronic oscillator.
• To enhance the performance of prediction and classification tasks compared to traditional RC methods.

Main Methods:

• A dual-loop optoelectronic oscillator was designed and implemented, requiring only a minimal addition of fiber optic components.
• The proposed system was simulated and evaluated on three benchmark tasks: nonlinear auto regressive moving average (NARMA10), signal waveform recognition, and handwritten numeral recognition.
• Parameter optimization for the NARMA10 task was conducted to identify influential factors.

Main Results:

• The novel RC system achieved a normalized mean square error of 0.0493±0.007 on the NARMA10 task.
• A signal waveform recognition task yielded an error of 6.172×10^-6.
• Handwritten numeral recognition resulted in a word error rate of 9% with optimized parameters.

Conclusions:

• The proposed dual-loop optoelectronic oscillator-based reservoir computing system offers a significant performance enhancement.
• The compact hardware design and improved accuracy make it a promising approach for advanced temporal information processing.