Cognitive flexibility versus stability via activation-based and weight-based adaptations

Shengjie Xu, Tom Verguts, Senne Braem

Abstract: Humans are remarkably efficient at adapting to different contextual demands by exerting optimal levels of cognitive flexibility versus stability for switching between different tasks. Here, we show how a recurrent neural network can be used to simulate behavioral indices of cognitive flexibility versus stability, and investigate how people learn and apply optimal control settings across different contexts. Our model showed both fast but transient adaptation (in activation space) to high- versus low-switch frequency conditions, and slow but more enduring shallow versus deep task attractor settings (in weight space) in these same contexts. Interestingly, it further learned to associate and use information from its context to different task attractor settings and shift along a flexibility-stability continuum. In sum, we provide a novel framework that sheds new light on classic measures of cognitive flexibility versus stability when people must switch between several tasks, through the lens of activation-based versus weight-based adaptations.