The human brain has a remarkable ability to store and retrieve information. Detailed memories can be formed after as little as one exposure, and those memories can be retained for decades. Importantly, recent studies have demonstrated the critical role of a mental “schema”, or a learned cognitive structure, in supporting rapid memory formation. In essence, previously acquired knowledge provides a framework that shapes how ongoing experience is perceived and remembered. This resonates with the older idea from neuropsychology of “learning to learn” as a mechanism by which rapid learning might be accomplished. Learning to learn is thought to facilitate new learning by reducing the dimensionality of the space that the organism has to search to adapt to novel problems. The development of a schema that supports this kind of rapid learning is thought to involve interactions between the hippocampus and the neocortex. However, the neural circuitry that underlies this kind of rapid, one-trial, learning is not well-understood.
Here, using innovative techniques for large-scale recordings in monkeys and in human epilepsy patients, we capitalize on the unique opportunity to establish cross-species comparisons of hippocampal-neocortical interactions. We record single unit activity as well as the local field potential from homologous brain areas in humans and monkeys, and we investigate this neural activity as humans and monkeys perform identical behavioral tasks of rapid learning and schema formation. In parallel, computational modeling and theoretical work provides an important iterative loop for evaluating our experimental hypotheses. Recently, advances in the modeling of recurrent neural networks provide a very promising framework within which to interpret the kind of complex data we will be gathering. By training artificial networks to perform the same tasks as our experimental subjects, these learned dynamics serve as a plausible network implementation of the cognitive computation. We can then examine these networks to discover the hidden dynamical structure of these computations, and to develop hypotheses for how the biological network may implement them. The overall goal of this U-19 Program is to develop a comprehensive theory of the circuit mechanisms that underlie the human brain’s ability to establish neural frameworks that enable rapid new learning.