Modeling and Analysis of Fish Interaction Networks under Projected Visual Stimuli

A team of researchers led by Hiroaki Kawashima has published a novel AI framework for analyzing collective animal behavior. Their paper, 'Modeling and Analysis of Fish Interaction Networks under Projected Visual Stimuli,' presents a method to estimate a dynamic 'influence network' within a group, such as a fish school, while accounting for external visual stimuli projected into their environment. Building on a prior model that used sparse regression to infer social networks from movement trajectories, this new formulation adds a dedicated stimulus term. This allows the model to disentangle and quantify two key forces: the influence individuals have on each other and how they collectively react to and propagate external visual cues.

The technical core combines a Boids-type simulation model—which mimics flocking behavior—with a machine learning framework for network inference. The key innovation is the simultaneous estimation of inter-individual interaction strengths and stimulus-related influence weights from observational data. The researchers also developed entropy-based indices to measure bias in an individual's influence, effectively identifying 'leader' fish within the school. In experiments, this framework successfully quantified schooling behavior and pinpointed influential individuals, creating a basis for real-time, interpretable metrics of group dynamics. This work, presented at AROB-ISBC 2026, bridges computational social network analysis, quantitative biology, and AI, offering a powerful new lens for studying complex systems.