Coherent optical neural networks that have optical-frequency-controlled behavior and generalization ability in the...

Area of Science:

• Optics and Photonics
• Artificial Intelligence
• Information Theory

Background:

• Coherent optical neural networks offer advanced computational capabilities.
• Optical frequency control presents a novel learning parameter for these systems.
• Understanding generalization in frequency space is key for practical applications.

Purpose of the Study:

• To propose and analyze coherent optical neural networks utilizing optical frequency as a learning parameter.
• To investigate the learning process and generalization abilities in the frequency domain.
• To determine optimal parameter ranges for stable learning and effective generalization.

Main Methods:

• Development of a coherent optical neural network system with a complex-valued network, phase reference, and self-homodyne detection.
• Utilizing optical frequency as a learning parameter to adjust connection delay and transparency.
• Analysis of information geometry to identify suitable parameter ranges for generalization.
• Conducting simulation experiments to validate theoretical findings.

Main Results:

• Identified periodic error-function minima in both delay-time and input-signal-frequency domains.
• Established that initial connection delay must be within a specific range for meaningful generalization.
• Demonstrated stable learning and reasonable generalization in the frequency domain through simulations.

Conclusions:

• Coherent optical neural networks with optical frequency control are viable sophisticated optical systems.
• The identified parameter ranges derived from information geometry enable stable learning and effective frequency-domain generalization.
• This work provides a theoretical and experimental foundation for designing advanced optical neural networks.