Neuronal mechanisms of learning and utilizing abstract representations - Project summary Using memories of past experiences to guide behavior in novel situations is an important function of the brain. Doing so efficiently is thought to rely on the process of abstraction, which extracts and summarizes relevant aspects from prior experiences and thereby enables usage of the acquired knowledge to solve novel future situations efficiently. Despite the importance of abstraction for human cognition, little is known about the underlying mechanisms and their disruption by disease or injury. The long-term goal of this research is a circuit-level understanding of abstraction to enable the development of new treatments for the devastating effects of cognitive disorders in which abstraction fails. Our experiments utilize the rare opportunity to record in-vivo from human single neurons simultaneously in multiple brain areas in patients undergoing treatment for drug resistant epilepsy. The overall objectives of this project are to build a multi-institutional (Cedars- Sinai/Caltech, UC Davis, Barrow Neurological Institute, Columbia University), integrated, and multi-disciplinary team to use human single neuron recordings to examine the neural substrate of abstraction. We will investigate three overarching hypotheses on the neural substrate of abstraction. First, we will determine the dynamics of learning abstract representations (Aim 1). Second, we will decipher how abstract representations facilitate generalization (Aim 2). Third, we will decipher how abstract representations facilitate analogical reasoning (Aim 3). The expected outcomes of this work are a characterization of how abstract neural representations are formed, how they enable generalization to novel stimuli and tasks, and how they enable analogical reasoning. This work is significant because we test circuit-based hypotheses by simultaneously recording single-neurons from multiple frontal cortical and subcortical temporal lobe areas in humans who are forming and using abstract representations. The proposed work is unusually innovative because we combine single-neuron recordings in in behaving humans with new tools to examine neural geometries and their relationship to behavior, develop new methods for the characterization of the format of neural information, test the major theory that it is abstract representations, and their property of compositionality, that enable out-of- distribution generalization, and we test the theory that abstract representations are build by disentangling existing representations much like learning occurs in machines. A second significant innovation is our team, which combines the patient volume and expertise of several teams to maximally utilize the rare neurosurgical opportunities available to study the human nervous system at single cell resolution. This innovative approach permits us to investigate circuit-level mechanisms of human cognition that cannot be studied non-invasively in humans nor in animal models. This integrated multi-disciplinary combination of physiology, behavior, and modeling will contribute significantly to our understanding of the circuits and patterns of neural activity that give rise to abstraction, which is a central goal of human neuroscience and the BRAIN initiative.
