Human learning is a continuous, multi-sensory, and social endeavor. Much of it is about extracting useful relational patterns across different situations, so we can adaptively act in the environment, including effectively acquiring and utilizing various human-invented symbol systems (e.g., language, numbers). These characteristics of human learning guide our approach in research. We combine large-scale cross-sectional and longitudinal studies with multi-session training experiments, computational modeling, high temporal resolution behavioral data collection (e.g., eye tracking), and translational research in schools. Our goal is to characterize the structure of the input data in children’s early learning environment, the processes and mechanisms through which children attend to, represent, and learn from this data, how this learning creates hidden deficits or competencies in children’s cognitive system, and how those changes do or do not prepare children for learning from formal instruction in schools.