Computing by Robust Transience: How the Fronto-Parietal Network Performs Sequential, Category-Based Decisions.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Science

Background:

• Decision-making integrates internal judgments with external perception, studied via delayed match-to-category (DMC) tasks.
• Neuronal recordings in LIP and PFC show mixed, time-varying selectivity during DMC tasks, but links to population computation remain unclear.

Purpose of the Study:

• To link observed neural selectivity characteristics to population-level computations in decision-making.
• To develop and analyze a recurrent network model that replicates neural and behavioral aspects of DMC tasks.

Main Methods:

• Training a recurrent neural network to perform DMC tasks.
• Analyzing single-neuron and population-level activity within the trained network.
• Investigating the role of network connectivity and emergent dynamics in decision formation.

Main Results:

• The recurrent network model successfully reproduced key features of neuronal selectivity observed in biological experiments.
• Robust transient trajectories of neural populations were identified as critical for sequential categorical decisions.
• Network self-organized connectivity shapes these trajectories, forming a
• neural landscape
• with slow states and dynamical tunnels.

Conclusions:

• Neural population dynamics, particularly transient trajectories within a structured
• neural landscape
• , are fundamental to sequential categorization.
• The developed model provides a framework for understanding decision-making mechanisms and can be generalized to other categorization tasks.