The role of population structure in computations through neural dynamics

Area of Science:

• Computational neuroscience
• Systems neuroscience
• Neural dynamics

Background:

• Current research separates neural population analysis into studying functional subpopulations or collective activity dynamics.
• The interaction between these two aspects in shaping neural computations remains poorly understood.

Purpose of the Study:

• To investigate the interplay between neural population structure and the dimensionality of collective activity in implementing computational tasks.
• To elucidate how these factors contribute to flexible and robust neural information processing.

Main Methods:

• Developed a novel computational approach to extract mechanisms from artificial neural networks trained on neuroscience-relevant tasks.
• Analyzed the relationship between network dimensionality, subpopulation structure, and task performance.
• Investigated the role of gain-controlled modulations in shaping collective neural dynamics.

Main Results:

• While high dimensionality can support various tasks in random networks, flexible input-output mapping necessitates non-random subpopulation structures.
• A defined subpopulation structure facilitates flexible computations via gain-controlled modulations that dynamically shape collective activity.
• The study demonstrates complementary roles for dynamics dimensionality and subpopulation organization.

Conclusions:

• Neural computations are shaped by a synergistic interplay between the dimensionality of collective activity and the organization of neural subpopulations.
• Subpopulation structure is crucial for enabling flexible computations, particularly for complex input-output mappings.
• Findings predict task-specific neural selectivity, inform inactivation experiments, and clarify neuron roles in multi-tasking.