Prof. Shangce Gao

Artificial intelligence (AI) has achieved remarkable advances, yet the growing scale of modern models brings substantial computational and energy costs. This presentation introduces research at the University of Toyama Artificial Intelligence Laboratory aimed at moving beyond “bigger AI” toward smarter, more efficient, and reliable AI systems. Our approach to Green AI draws inspiration from the information-processing capabilities of biological dendrites. By developing dendritic neuron models and learning methods, we seek compact, interpretable network structures that can deliver strong performance with reduced computational demands.

The talk highlights representative applications of this research philosophy. A lightweight multi-dendritic model demonstrates competitive image recognition performance with a substantially reduced model size. In medical AI, PCNet addresses the segmentation of tiny anatomical structures, while a fully complex-valued gated recurrent neural network enables fast, high-quality ultrasound image reconstruction. Further examples include turbulent pressure prediction from limited observations and long-term time-series forecasting for applications such as energy, transportation, and weather. Together, these studies illustrate how carefully designed AI can support practical challenges across medicine, engineering, and society.

Looking ahead, we envision AI systems that continue learning, adapt autonomously to changing environments, and maintain reliable performance over time. The central message is that the next generation of AI should be judged not by model size alone, but by the combination of high performance, high efficiency, and high reliability.

Keywords: Green AI; dendritic learning; lightweight neural networks; medical AI; AI for Science

Image

Prof. Shangce Gao

School of Engineering, University of Toyama