A model of working memory for encoding multiple items and ordered sequences exploiting the theta-gamma code

Area of Science:

• Computational Neuroscience
• Cognitive Neuroscience
• Artificial Intelligence

Background:

• Oscillatory brain activity, particularly theta and gamma rhythms, is crucial for working memory maintenance in humans and rodents.
• Cross-frequency coupling between theta and gamma oscillations is hypothesized as a key mechanism for managing multiple memory items.

Purpose of the Study:

• To present an original neural network model based on oscillating neural masses to investigate working memory mechanisms.
• To explore how the model handles single-item reconstruction, multi-item maintenance without order, and ordered sequence recall.

Main Methods:

• Developed a four-layered neural network model utilizing oscillating neural masses.
• Employed Hebbian and anti-Hebbian learning rules for synapse training to synchronize/desynchronize features.
• Simulated network performance under various conditions, including altered synaptic strengths and an 'imagination phase'.

Main Results:

• The model successfully desynchronized up to nine items using gamma oscillations, demonstrating order-independent multi-item maintenance.
• Sequences were replicated by nesting gamma rhythms within theta rhythms.
• Reduced GABAergic synapse strength mimicked neurological memory deficits.
• The 'imagination phase' allowed random sequence recovery and linking based on item similarity.

Conclusions:

• The proposed neural network model provides a viable framework for understanding working memory dynamics.
• The model's ability to simulate multi-item storage, sequence recall, and memory alterations highlights the role of oscillatory activity and synaptic plasticity.