Search Papers | Poster Sessions | All Posters
Poster B54 in Poster Session B - Thursday, August 8, 2024, 1:30 – 3:30 pm, Johnson Ice Rink
Dynamic neural control of task representations in humans and neural networks
Harrison Ritz1, Aditi Jha, Jonathan Pillow, Nathaniel Daw, Jonathan D Cohen; 1Princeton University
Cognitive psychologists often model task preparation using dynamical systems theory, however the neural correlates of these cognitive dynamics remain poorly understood. We bridged between cognitive and neural theories by fitting linear dynamical systems to human EEG recordings during task switching. Using a control theoretic analysis of the fitted dynamical system, we found that participants showed stronger propagation of task information when switching tasks than when repeating tasks. Similar signatures of task control were evident in task-optimized neural networks, consistent with this neural marker of task reconfiguration reflecting an optimality principle.
Keywords: Task Switching EEG Latent State Space Recurrent Neural Network