Abstract
Predictive coding has emerged as a compelling framework for understanding perception and action in biological and artificial agents. In this work, we explore the implications of predictive coding in a multi-agent system, where each agent moves in a continuous environment and continuously predicts its future state based on sensory inputs. Agents minimize their prediction error by adapting their movement, leading to emergent collective behaviors such as aggregation, cohesion, and obstacle avoidance. Through simulations, we investigate how these behaviors arise from local predictive mechanisms without the need for explicit coordination or predefined interaction rules. Our results show that predictive coding not only enables individual agents to navigate their environment efficiently but also fosters self-organized collective structures that resemble swarm-like behavior observed in natural systems. The findings suggest that predictive coding can serve as a powerful principle for designing distributed intelligent systems capable of adaptive and emergent coordination.
About the speaker
Nicolas Bessone is a PhD student at the IT University of Denmark, focusing on decentralized controllers for distributed systems and a multitude of side projects that range from agent-based modeling and music theory representations in computational spaces to neuroscience, physics, and artificial intelligence.