CRCNS Research Proposal: Attentional Control of Credit Assignment in Learning

A hallmark of human intelligence is our ability to flexibly adapt how we learn. Yet, the brain mechanisms underlying this flexibility remain poorly understood. A better understanding of adaptive learning strategies in humans can inform approaches to understand and aid people with learning disorders, as well as guide development of artificial systems that learn in more human-like ways. In cognitive science and artificial intelligence research, a fundamental challenge in learning is the process of credit assignment, which describes how a biological or artificial agent generalizes from past experiences to future conditions in complex environments. This project examines the computational and neural bases of how humans adapt their credit assignment strategies during learning. A key hypothesis is that the psychological and neural processes of selective attention are engaged to flexibly control humans’ credit assignment strategies according to demands of the environment. In addition to the research, this project includes interdisciplinary training of junior scientists in computational and cognitive neuroscience, as well as science education outreach to K-12 students. This project develops and tests computational models of adaptive attentional control of credit assignment, in conjunction with experimental characterization of behavior and brain activity in humans performing learning multi-feature reinforcement learning tasks. Choice behavior on these tasks are fit with kernel-based models, to test the adaptation of credit assignment strategies to task demands. Eye-tracking measurements characterize the dynamics of gaze, to characterize how gaze relates to choice behavior and to test how deployment of spatial attention supports flexible weighting of stimulus features for decision making and credit assignment. Noninvasive electrophysiology of brain activity with EEG characterizes how established neural correlates of spatial attention track behavioral weighting of features, to test the neural mechanisms through which attention can shape credit assignment. Informed by these experimental studies, computational modeling of learning in artificial neural networks, integrating attentional mechanisms, generates behavioral and neural predictions and yield theoretical insights into the interplay of attention and learning in the human brain. This project’s combination of computational and experimental research bridges the fields of computational and cognitive neuroscience with artificial intelligence to advance our understanding of human learning. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.