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

Area of Science:

• Cognitive Neuroscience
• Computational Neuroscience
• Artificial Intelligence

Background:

• Human cognitive control involves balancing flexibility and stability for task switching.
• Control adaptation occurs across multiple timescales, suggesting diverse underlying mechanisms.

Purpose of the Study:

• To develop a recurrent neural network model simulating cognitive flexibility and stability dynamics.
• To investigate the computational mechanisms behind fast (activation-based) and slow (weight-based) control adaptations.

Main Methods:

• Developed a recurrent neural network model to simulate behavioral indices of cognitive control.
• Conducted four simulation studies to analyze model dynamics.
• Validated model predictions using an existing human dataset (N=102).

Main Results:

• Activation-based adaptations facilitate rapid shifts in cognitive flexibility.
• Weight-based adaptations, dependent on control settings, exhibit slower learning dynamics.
• The model learned to associate context with control settings, mimicking human context-dependent flexibility-stability shifts.

Conclusions:

• Integrated fast and slow control adaptations into a unified computational framework.
• Advanced understanding of human adaptive behavior and regulatory mechanisms.
• Highlighted the distinct roles of activation-based and weight-based adaptations in cognitive control.