A hybrid biological neural network model for solving problems in cognitive planning

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Science

Background:

• Behaviors like spatial navigation and bodily motion can be modeled as graph traversal problems using cognitive maps.
• Existing models may not fully align with empirical findings from the mammalian neocortex and hippocampus.

Purpose of the Study:

• To present a novel neural network model capable of solving graph traversal tasks.
• To ensure the model is compatible with established findings in mammalian neurobiology, specifically the neocortex and hippocampus.

Main Methods:

• A neural network model is proposed where neurons and synaptic connections self-organize into a cognitive map via Hebbian learning.
• The network encodes a distance metric on an abstract task-relevant manifold, with neurons representing points on this manifold.
• Graph traversal is achieved through wave-like activation patterns guiding activity peaks from start to target states.

Main Results:

• The model successfully formulates behaviors as graph traversal problems.
• It demonstrates compatibility with a broad range of empirical findings regarding the mammalian neocortex and hippocampus.
• Wave-like activation patterns effectively guide localized activity peaks for pathfinding.

Conclusions:

• The developed neural network model offers a biologically plausible mechanism for solving complex navigation and motion tasks.
• It provides a framework for understanding how self-organized cognitive maps in the brain support goal-directed behaviors.
• The model's compatibility with neurobiological data suggests its potential for further research in artificial intelligence and neuroscience.