On-chip bacterial foraging training in silicon photonic circuits for projection-enabled nonlinear classification

Area of Science:

• Photonics
• Machine Learning
• Silicon Photonics

Background:

• On-chip training for photonic devices is a significant challenge.
• Current photonic machine learning primarily focuses on inference for pre-trained models.
• Support Vector Machines (SVM) are under-explored in photonic implementations.

Purpose of the Study:

• To propose and demonstrate a projection-based classification principle in silicon photonic circuits.
• To implement on-chip training using natural-intelligence-inspired bacterial foraging.
• To explore the potential of SVM-inspired methods in photonic machine learning.

Main Methods:

• Constructing nonlinear mapping functions in silicon photonic circuits.
• Utilizing bacterial foraging for on-chip training.
• Experimentally demonstrating classification for Boolean logic and Iris dataset.

Main Results:

• Achieved high classification accuracy (~96.7 - 98.3%) for single Boolean logics, combinational Boolean logics, and Iris classification.
• Demonstrated comparable performance to artificial neural networks with smaller-scale photonic circuits.
• Showcased scalability advantages without traditional activation functions.

Conclusions:

• This work enables photonic circuits to perform nonlinear classification tasks with on-chip training.
• Bacterial foraging provides an efficient and robust training method for photonic devices.
• The proposed approach offers a promising pathway for advanced machine learning in photonics.