Keynotes | K&Ts | GACs | Talks | Posters | Search

Poster C in Poster Session C: Wednesday, August 5, 9:30 – 11:15 am, Kimmel Center, Shorin & Rosenthal Rooms

Representational Structure of Ordered Contexts

Siddhant Iyer¹, Rhiannon L Cowan², Shervin Rahimpour¹, Stefano Fusi¹, Ben Shofty¹, Elliot H Smith², Nuttida Rungratsameetaweemana¹; ¹Columbia University, ²University of Utah

Presenter: Siddhant Iyer

Animals and humans adjust their behavior depending on the context, which is supported by a restructuring of neuronal population activity (representation) in the Prefrontal Cortex (PFC) and Hippocampus (HIPP). To efficiently switch between contexts, the brain must know how their representations are structured relative to one another. For example, across casual, business-casual, or business events (contexts), the appropriate attire choice (behavior) increases in its formality. Here, we specifically ask how the brain organizes such ordered contexts. We designed a categorization task, where the decision boundary (rule) is shifted to create three contexts. Human participants (n=6) and task-trained neural networks (n=30) exhibited behavioral shifts along this context space, also reflected in the networks' hidden layer activities. Accurate context-dependent readout of behavior required de-shifting representations along context space. Overall, we show that neural networks can flexibly switch between ordered contexts by shifting their readouts along context space. Next steps involve validation of these neural predictions on human single-neuron recordings from the PFC and HIPP of neurosurgical patients performing the task.

Topic Area: Decision-Making, Cognitive Control & Event Cognition