Probabilistically segregated neural circuits and subcritical linguistics

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Science

Background:

• Early models of cortical information processing and memory capacity relied on large neural networks with binary (on/off) activity.
• These models, under the Hebbian paradigm, suggested high storage and retrieval capacities but resulted in unrealistic, complex code words due to high cross-network connectivity.

Purpose of the Study:

• To investigate the limitations of large neural networks in representing realistic cortical information codes.
• To explore how neural circuit activity states (polarities) influence network structure and coding efficiency.
• To demonstrate that specific network configurations can lead to more plausible and memorable information representations.

Main Methods:

• Analysis of neural network connectivity and code word length under different activity assumptions.
• Application of random-graph theory to understand neural circuit segregation based on polarity probability.
• Evaluation of the linguistic plausibility and memorability of code words generated by segregated circuits.

Main Results:

• Large, evenly active neural networks produce long, linguistically implausible code words, hindering comprehension and memory.
• Neural circuit activity is jointly governed by somatic and synaptic states (neural circuit polarities).
• Subcritical polarity probability leads to small neural circuit segregation, generating linguistically plausible code words.

Conclusions:

• Segregated neural circuits, arising from specific polarity dynamics, are crucial for efficient and realistic cortical information coding.
• These findings suggest a more biologically plausible mechanism for memory storage and retrieval than previously assumed.
• The study highlights the importance of network structure and activity states in shaping cognitive functions like memory.