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

Poster E in Poster Session E: Thursday, August 6, 10:30 am – 12:15 pm, Kimmel Center, Shorin & Rosenthal Rooms

Towards Inference of Hidden Causes with Assemblies

Max Dabagia¹, Christos Papadimitriou²; ¹Columbia University, ²University of California-Berkeley

Presenter: Max Dabagia

The cortex is a formidable learning engine, but what does it learn? One hypothesis is that it creates a hidden Markov model of its input, to perform probabilistic inference about the hidden causes of the input stimuli. However, prior work focused on sequences of inputs with precise timing and occurring in quick succession, with stimulus events represented by single spikes. When inputs arrive more slowly and with realistic variability in timing, meaningful structure is often lost. We explore how assemblies of neurons can overcome this issue in a cortical network model. We show that, under certain STDP rules, distinct assemblies form to represent a given stimulus depending on other recently-seen stimuli. Crucially, these assemblies are a persistent form of short-term memory, capable of keeping themselves active indefinitely until a new stimulus arrives and then biasing the neural response to that stimulus, thus bridging the gap between neural and behavioral events. We apply this model to stimuli generated by movement along a linear track where instantaneous stimuli do not disambiguate position, and observe the emergence of assemblies which encode specific locations, necessarily using context.

Topic Area: Methods, Tools, Theory & Neural Coding