Combining Optimal Path Search With Task-Dependent Learning in a Neural Network.

Area of Science:

• Artificial Intelligence
• Computational Neuroscience
• Graph Theory

Background:

• Optimal path finding in graphs typically uses predefined edge costs, limiting adaptive planning.
• Conventional methods struggle with dynamic cost adjustments required by specific tasks.

Purpose of the Study:

• To present a neural network model for path-finding problems with adaptive cost adjustments.
• To demonstrate how neural network learning mechanisms can augment paths based on task requirements.

Main Methods:

• Representing path-finding problems using a neural network where edge costs are synaptic weights.
• Utilizing activity propagation for path computation, analogous to the Bellman-Ford algorithm.
• Employing network learning mechanisms, such as Hebbian learning, for weight adaptation.

Main Results:

• The neural network's activity propagation yields solutions identical to the Bellman-Ford algorithm.
• The network demonstrates adaptive path augmentation for tasks like obstacle navigation and sequence following.
• The proposed algorithm integrates path finding and learning.

Conclusions:

• A novel neural network approach enables adaptive path finding by learning synaptic weight adjustments.
• This method offers a new paradigm for applications requiring dynamic path optimization.
• The integration of learning with path finding opens diverse application possibilities.