Humans parsimoniously represent auditory sequences by pruning and completing the underlying network structure.

Area of Science:

• Cognitive Neuroscience
• Auditory Perception
• Network Science

Background:

• Auditory sequences contain dependencies at multiple scales, from local transitions to hierarchical structures.
• Human ability to learn these dependencies from limited data is not fully understood.
• Existing models often struggle to unify learning across different temporal scales.

Purpose of the Study:

• To investigate how humans represent local and higher-order structures in auditory sequences.
• To explore the interaction between different scale levels in auditory sequence learning.
• To model the cognitive and neural mechanisms underlying multi-scale auditory learning.

Main Methods:

• Utilized network science formalisms to analyze auditory sequences.
• Assessed human adult perception of local transitions and higher-order network structures.
• Developed an analytical model based on a memory/efficiency trade-off.

Main Results:

• Human adults show perceptual biases towards local transitions, enhancing sensitivity to high-order network structures like communities.
• This behavior aligns with creating a parsimonious, simplified model of auditory evidence by pruning and completing relationships.
• The proposed memory/efficiency trade-off model successfully explains both local transition probabilities and high-order structures.

Conclusions:

• The brain employs a parsimonious representation strategy, not relying on exact memories.
• A unified model of sequence learning across scales is achieved through a memory/efficiency trade-off.
• Putative neural implementations for this bias are proposed.