2023 Theses Doctoral

Towards a Neuroscience of Stories: Metric Space Learning in the Hippocampus

Liao, Zhenrui

The ability to recall, structure, and reason about our past experience is essential to both our functioning in the world, and our identities as humans. Memory is not solely a task of reactivating past experiences verbatim, but of constructing models (“stories”) to understand them. Rodent studies have allowed us extraordinary real-time access to the neural substrates of memory in the hippocampus, but thus far it has been difficult to ask rodents to perform the kind of flexible learning of complex worlds that characterizes our subjective experience of episodic memory.

In this work, I examine three levels of learning models of the world: (1) the dendritic input space of single cells from which novel concepts can be carved out, (2) the gradual learning of latent sequences of these concepts through experience in the presence of noise, and (3) the contextualization of these concepts and sequences in an incrementally constructed model of the world.

I utilize a multimodal approach combining two-photon voltage imaging data from single cells in vivo, spiking network models at different levels of abstraction, and two-photon population imaging in a novel virtual reality (VR) task to examine learning of non-Euclidean relational structures. Finally, I argue that these three components are sufficient to construct metric space-like representations in the hippocampus which can be used to solve relational queries on diverse spaces, with implications for how we acquire understanding of the world around us.