3/1/24 – Friday Neuroscience Brown Bag Lecture Series: Ken Rahman
Macaque learning of spatiotemporal sequences and their mapping to hippocampal representational geometry
Date: Friday, March 1
Time: 4:00 p.m. – 5:00 p.m.
Location: Wilson Hall 115
Hippocampal activity in freely-moving rodents shows the capacity for stable spatiotemporal representations of the environment (through ‘place cells’) that may serve as a cognitive map. These findings have been formative for developing models of abstraction in the brain and in AI. Yet it remains unclear how the place cells of rodents in mazes translate to abstraction of cognitive maps in humans, due in part to a paucity of evidence of the relevant representational spaces among neural ensembles in the primate hippocampus. To assess these representational spaces and their modification across structured experience, we first developed a quasi-naturalistic learning paradigm for macaques. In it, the monkeys learned multiple item-context sequences that were presented in blocks which alternated between opposite corners of the environment, thus requiring flexible learning in 3D space. Monkeys’ behavior was sensitive to recent and remotely learned sequences, developing performance strategies gradually over time. To Assess the neural representational spaces associated with this task, we recorded ensembles of single units as monkeys performed the task blocks, and applied unsupervised (UMAP) and contrastive supervised (CEBRA) dimensionality-reduction methods on neural ensemble activity from multiple sessions in two monkeys (per session yields were ~40-173 hippocampal units ; ~40-273 total per session). Despite stable unit recordings, UMAP revealed a drift in the low-dimensional space and an increase in the state-space coverage across trials. To better understand the task elements responsible, we used multi-session CEBRA, and found that the task blocks were decodable across the two sequences (CEBRA fl: 0.756618), and that the trial sequence components could be separated, with reward epochs showing the most exaggerated separation. Furthermore, we found that in some indicative of drift, or at least, of temporal segregation. This suggests that both temporal drift and the underlying structure of the sequences are reflected in the ensemble activity. Building on these initial results, I hope to clarify the most important dimensions onto which cognitive maps may be constructed in the macaque hippocampus.
Questions? Contact Kari Hoffman.