Practice Reshapes the Geometry and Dynamics of Task-tailored Representations.

Area of Science:

• Cognitive Neuroscience
• Computational Neuroscience

Background:

• Extensive practice leads to automaticity, improving task efficiency and precision.
• Theories of automaticity debate which representational levels (low-level features, stimulus-response, or high-level memories) change with practice.
• Neural population dynamics offer a framework to understand computational changes during learning.

Purpose of the Study:

• To investigate how neural representations change with extensive practice.
• To test the hypothesis that practice optimizes representational geometry for task contingencies.
• To determine which representational levels predict performance improvements.

Main Methods:

• Human participants (n=40) practiced a context-dependent action selection task over 3 days.
• Electroencephalography (EEG) recorded neural activity during practice.
• Analysis focused on changes in neural representations of task features and conjunctions.

Main Results:

• Neural representations of high-level, context-specific conjunctions were enhanced during rapid performance improvement.
• Enhancement of conjunctive representations, not lower-order ones, predicted power-law performance improvement.
• Conjunctive neural states became more stable and aligned, abstracting over redundant features.

Conclusions:

• Practice optimizes neural representational geometry, creating task-tailored states.
• This optimization involves enhancing high-level conjunctive representations.
• Improved representational dynamics explain performance gains and reduced switch costs.