Elucidating Infant Learning Principles Based on Predictive Coding

Human infants acquire the structure of the world and the intentions of others remarkably quickly from limited experience. This project aims to formulate the principles of infant learning as computational models grounded in predictive coding and active inference.

We model the integration of multimodal sensory information and the generative processes of expectations and prediction errors in social interactions with caregivers. From the perspective of cognitive developmental robotics, we bridge real-robot experiments and developmental psychology experiments. By describing infant learning principles as a reproducible computational theory, we aim to apply these insights to the understanding of developmental disorders and to early-learning algorithms for robots.