The neural network approach to a parallel decentralized network routing

Area of Science:

• Computer Science
• Network Engineering
• Artificial Intelligence

Background:

• High-speed optical transmission and packet switching necessitate efficient network routing solutions.
• Existing centralized routing methods, often based on large Hopfield-type neural networks, face scalability and control limitations.

Purpose of the Study:

• To present a novel parallel decentralized network routing method.
• To address the limitations of centralized control in large-capacity communication networks.
• To enable real-time, sub-optimum routing solutions.

Main Methods:

• A parallel decentralized model comprising interconnected, fully connected intraconnected networks at each node.
• Utilizing unique state equations for neuron dynamics to map to network routing problems.
• Leveraging high-speed convergence properties of neuron dynamics.

Main Results:

• The proposed method demonstrates effective network routing through decentralized control.
• Simulation results validate the feasibility and performance of the parallel decentralized approach.
• The model achieves sub-optimum routing solutions suitable for real-time applications.

Conclusions:

• The parallel decentralized network routing method offers a viable alternative to centralized approaches.
• The model's neuron dynamics facilitate efficient and rapid routing decisions.
• This approach is well-suited for the demands of modern high-capacity multimedia communication networks.