Effective learning is accompanied by high-dimensional and efficient representations of neural activity

Area of Science:

• Cognitive Neuroscience
• Neuroimaging
• Computational Neuroscience

Background:

• Mapping object value and identity is a core cognitive function.
• The optimal neural space for this value-identity mapping remains unclear.
• Understanding this mapping is crucial for deciphering learning mechanisms.

Purpose of the Study:

• To develop quantitative tools for analyzing neural representations of value-identity mapping.
• To investigate how learning speed influences the dimensionality and organization of neural responses.
• To identify geometric measures for assessing cognitive coding efficiency.

Main Methods:

• Utilized functional magnetic resonance imaging (fMRI) to capture neural responses.
• Developed novel methods to quantify the dimensional space and organization of neural mappings.
• Analyzed whole-brain responses to objects varying in value.
• Investigated neurophysiological drivers at smaller scales.

Main Results:

• Quick learners demonstrated higher-dimensional neural representations compared to slow learners.
• Neural responses in quick learners were more distinguishable across the whole brain.
• Quick learners showed more compact neural embeddings, indicating efficient cognitive coding.
• Higher ratios of stimuli dimension to embedding dimension were observed in quick learners.

Conclusions:

• Neural response organization is characteristic of learning efficiency.
• Geometric measures can quantify and identify efficient coding in higher-order cognition.
• The study provides insights into the neural basis of value-based learning and representation.