Neural computing with coherent laser networks

Area of Science:

• Physics
• Computer Science
• Optical Engineering

Background:

• Coherent laser networks (CLNs) are explored for their potential in computational tasks.
• Traditional neural networks often struggle with continuous data processing.

Purpose of the Study:

• To demonstrate emergent neural computing capabilities in CLNs.
• To investigate pattern storage and retrieval mechanisms within CLNs.
• To overcome limitations in CLN storage capacity and pattern retrieval.

Main Methods:

• Harnessing the collective behavior of laser networks to store phase patterns.
• Analyzing dynamical equations to identify stable fixed points representing stored information.
• Implementing nonreciprocal coupling to improve pattern retrieval fidelity.

Main Results:

• CLNs exhibit associative memory properties by storing and retrieving phase patterns.
• Large overlap between fixed-point patterns initially limits retrieval to two images.
• Nonreciprocal coupling successfully overcomes this limitation, enabling utilization of the network's full storage capacity.

Conclusions:

• CLNs offer novel possibilities for neural computation as analog processors.
• The dynamical model suggests a new energy-based recurrent neural network for continuous data.
• This research bridges optical systems and advanced neural network architectures.