Photonic Kolmogorov-Arnold networks based on self-phase modulation in nonlinear waveguides

Area of Science:

• Optics and Photonics
• Artificial Intelligence
• Nonlinear Optics

Background:

• Photonic neural networks offer low-power, low-latency, and highly parallelized hardware.
• Kolmogorov-Arnold networks present improved scalability and interpretability over traditional models.

Purpose of the Study:

• To propose and experimentally demonstrate a photonic implementation of Kolmogorov-Arnold networks.
• To utilize self-phase modulation in nonlinear waveguides for generating activation functions.

Main Methods:

• Proposed using self-phase modulation in nonlinear waveguides.
• Experimentally showcased the generation of arbitrary nonlinear activation functions.
• Implemented and tested on regression and classification tasks.

Main Results:

• Photonic implementation of Kolmogorov-Arnold networks achieved competitive performance.
• Demonstrated ultra-fast activation function generation.
• Outperformed digital baselines in specific regression and classification tasks.

Conclusions:

• Self-phase modulation in nonlinear waveguides is a versatile method for photonic neural networks.
• This approach overcomes scalability barriers in photonic implementations.
• Enables efficient and high-performance photonic Kolmogorov-Arnold networks.