The information bottleneck as a principle underlying multi-area cortical representations during decision-making

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Science

Background:

• Decision-making involves distributed brain computations across multiple areas.
• The functional advantage of distributing computations in the brain remains unclear.
• Artificial neural networks utilize multi-layered architectures to form optimal task representations.

Purpose of the Study:

• To investigate why the brain distributes computations for decision-making.
• To identify how cortical areas form minimal sufficient representations.
• To explore the neural mechanisms underlying optimal information processing in decision-making tasks.

Main Methods:

• Recorded single neurons and multiunits in dorsolateral prefrontal cortex (DLPFC) and dorsal premotor cortex (PMd) in monkeys during a perceptual decision-making task.
• Trained a multi-area recurrent neural network (RNN) to perform the same task.
• Analyzed neural representations and inter-area communication patterns.

Main Results:

• DLPFC represents task inputs, while PMd contains a minimal sufficient representation of the choice.
• A multi-area RNN developed representations analogous to DLPFC and PMd.
• DLPFC-like areas orthogonalized choice information and preferentially propagated it downstream via selective connections.

Conclusions:

• Cortical computation across multiple areas generates minimal sufficient representations for decision-making.
• Preferential propagation of relevant information between areas is a key mechanism.
• This multi-area strategy optimizes information processing for cognitive tasks.