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

On-chip training remains a challenging issue for photonic devices to implement machine learning algorithms. This paper proposes a projection-based classification principle using nonlinear mapping functions in silicon photonic circuits, experimentally demonstrating on-chip bacterial foraging training for single Boolean logics, combinational Boolean logics, and Iris classification with high accuracy (~96.7–98.3%). This approach offers comparable performance to artificial neural networks with smaller scales and without activation functions, showcasing scalability advantages. Bacterial foraging optimization provides efficient and robust on-chip training, paving the way for photonic circuits to perform nonlinear classification.

Guangwei Cong, Noritsugu Yamamoto, Takashi Inoue, Yuriko Maegami, Morifumi Ohno, Shota Kita, Shu Namiki, Koji Yamada