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

Poster F in Poster Session F: Thursday, August 6, 1:45 – 3:30 pm, Kimmel Center, Shorin & Rosenthal Rooms

How the Brain Can Build on Prior Learning: A Model and a Supporting Experiment

Lillian Chang¹, Alexander Farahbod¹, Andrew Kim¹, Christopher Cornell¹, Noah Steinberg¹, Florencia Martinez Addiego², Yuri Jung¹, Drew Di Donna¹, Tyler Morgan³, Peter Bandettini³, Maximilian Riesenhuber¹; ¹Georgetown University, ²Georgia Institute of Technology, ³National Institute of Mental Health

Presenter: Lillian Chang

It is currently not well understood how the brain builds on prior learning to facilitate the learning of novel tasks. This project tests a novel mechanistic model: after the prefrontal cortex (PFC) initially learns a new task, it provides re-entrant teaching signals to the anterior temporal lobe (ATL). This process "offloads" learned task circuits, making them potentially available as building blocks for future learning of novel tasks (as, e.g., in “fast mapping”). Notably, the learning algorithm does not require the backpropagation of errors, for which there is little empirical evidence. This model predicts that when learning a novel task, initial learning happens in PFC, then increased top-down information flow from PFC to temporal cortex after the task has been learned signals the transfer of the learned task module, indicated by increased category selectivity in temporal cortex and decreased information flow from frontal to temporal cortex after the task circuitry has been transferred. The model is supported by computational simulations as well as EEG data.

Topic Area: Memory, Learning & Knowledge Structures