University of Tokyo researchers develop optical AI framework to cut computation needs

The team, led by Associate Professor Kohei Nakajima and graduate student Yongbo Zhang, combined a training method called SS-DFA with a simpler network architecture called the single-map framework. This design avoids the need for extensive intermediate measurements during training, cutting down on data storage and computation. Tests on an optical computing system showed the framework matched or outperformed conventional neural networks while requiring far less processing power. Future work will explore scaling this to larger AI models and integrating it into consumer devices like phones or self-driving cars.