Distributed representations of action sequences in anterior cingulate cortex: A recurrent neural network approach

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Science

Background:

• The precise computational function of the anterior cingulate cortex (ACC) remains debated despite extensive research.
• Existing models struggle to fully capture the ACC's role in complex behaviors.

Purpose of the Study:

• To develop and test a computational model of the ACC based on its proposed function in executing goal-directed action sequences.
• To investigate how distributed representations in a neural network can explain ACC activity patterns.

Main Methods:

• A recurrent neural network was trained to predict successive steps in various goal-directed action sequences.
• The model incorporated distributed representations across interconnected processing units.
• Simulations were designed to mimic neurophysiological and neuroimaging observations.

Main Results:

• The network exhibited distributed activity patterns across simulated ACC neurons, mirroring sequence progression.
• The model generated discrepancy signals when action sequence steps deviated from predictions, consistent with human neuroimaging data.
• Simulated neurophysiological observations from nonhuman animals were replicated.

Conclusions:

• The computational model provides a plausible explanation for ACC function in managing extended, goal-directed action sequences.
• The findings suggest that distributed representations are key to understanding ACC's role in sequence execution and error detection.
• This modeling approach offers a novel framework for future investigations into ACC function.