Learning of Chunking Sequences in Cognition and Behavior.

Area of Science:

• Cognitive Neuroscience
• Computational Neuroscience
• Dynamical Systems Theory

Background:

• Humans and animals segment long sequences (e.g., phone numbers) into smaller chunks for learning and recall.
• The underlying dynamical principles governing this chunking behavior remain largely unknown.

Purpose of the Study:

• To investigate the temporal dynamics of sequence chunking using a novel dynamical model.
• To elucidate the mechanisms by which cognitive sequences are learned and robustly recalled through chunking.

Main Methods:

• Development of a dynamical model based on competing modes and hierarchical Winnerless Competition (WLC).
• Representation of sequential memory as trajectories along metastable fixed points.
• Utilizing bistable Hebbian dynamics for unsupervised learning of sequence trajectories.

Main Results:

• Demonstration of successful learning and robust recall of chunking representations for sequences via computer simulations.
• Association of modes with sequence items and encoding of their relative order during learning.
• Hierarchical WLC ensures sequence order robustness for shorter sequences.

Conclusions:

• The model's activity patterns align with behavioral observations, including chunk boundary pauses, size, and duration.
• Failures in learning chunking sequences may offer insights into the dynamics of neurological disorders like Parkinson's disease and Schizophrenia.