Topology reconfigurable photonic reservoir computing with multiple optical feedback loops

Area of Science:

• Optics and Photonics
• Machine Learning
• Complex Systems

Background:

• Conventional photonic time-delay reservoir computing (TDRC) utilizes a fixed ring topology, limiting its performance on tasks with varying dynamic requirements.
• Adaptability in TDRC is crucial for optimizing computational performance across different applications.

Purpose of the Study:

• To propose and demonstrate a task-dependent topology tuning mechanism for photonic TDRC.
• To enable hardware-level reconfiguration of the virtual network topology in TDRC using multiple optical feedback loops.

Main Methods:

• Implementing multiple optical feedback loops to create reconfigurable time-multiplexed virtual networks.
• Configuring delay parameters at the hardware level to tune the network topology.
• Evaluating performance on chaotic time series prediction and spatial classification tasks.

Main Results:

• Task-dependent topology tuning significantly impacts TDRC performance.
• Irregular nonresonant delays improve chaotic time series prediction by increasing state dimension and memory.
• Delay values synchronized with input scanning enhance spatial classification by recovering 2D correlations.

Conclusions:

• Multiple optical feedback loops offer a practical physical method for topology reconfiguration in photonic TDRC.
• Hardware-level topology tuning enhances the adaptability and performance of photonic reservoir computing systems.
• The optimal topology is strongly dependent on the specific computational task.