Learners are highly sensitive to pairwise statistical associations embedded in sensory input (e.g., what is the probability one element will follow another in time?). However, it remains an essential question how we use this information to build up complex knowledge systems (e.g., language), particularly in the face of noise or competing signals. Drawing on insights from functional neuroimaging, I will discuss the interplay between high-level association areas and sensory-specific cortex in a dynamic learning context. I will show that prefrontal cortex, a slow-to-mature area associated with cognitive control, underpins sequential pattern learning in adults, raising the possibility that they recruit a sub-optimal learning system relative to children. Through a series of behavioral experiments, I will then demonstrate that tools from network science offer a novel and largely untapped means of probing how learners scale up pairwise associations to gain knowledge of broad-scale patterns in their environment.