Lateral thinking, from the Hopfield model to cortical dynamics.

Area of Science:

• Computational Neuroscience
• Cognitive Neuroscience
• Systems Neuroscience

Background:

• Self-organizing attractor networks are proposed as fundamental units for cortical dynamics.
• These networks perform basic categorization operations, including analog-to-digital conversion and association.
• The expression of these operations in cognitive functions depends on neural code content.

Purpose of the Study:

• To assess the viability of modeling cortical dynamics using self-organizing attractor networks.
• To explore how memory representations self-organize in attractor networks, unlike artificial storage methods.
• To investigate the potential of network models for simulating spontaneous lateral thought.

Main Methods:

• Modeling local cortical patches as attractor networks with self-organizing, continuously graded patterns.
• Analyzing macaque recordings for evidence of rapid retrieval dynamics in cortical attractor networks.
• Employing a Potts model to capture large-scale cortical networks with adaptation dynamics.

Main Results:

• Cortical attractor networks can self-organize memory representations from afferent input, not as artificial binary patterns.
• Observed retrieval dynamics in macaques occur on rapid, cognitively plausible timescales.
• The Potts model demonstrates capacity for long-range associations and sustained attractor dynamics.

Conclusions:

• Self-organizing attractor networks provide a viable framework for understanding cortical dynamics and cognitive functions.
• These networks support rapid memory retrieval, potentially underlying cognitive processes.
• Network models, like the Potts model, can mathematically represent spontaneous lateral thought at a cortical scale.