A biologically inspired neural network for dynamic programming

Area of Science:

• Artificial Intelligence
• Computational Neuroscience
• Operations Research

Background:

• Dynamic optimization problems often present significant computational challenges.
• Conventional methods for dynamic programming can suffer from computational complexity.
• Artificial neural networks offer potential for parallel processing and efficient computation.

Purpose of the Study:

• To develop a novel artificial neural network for solving nonlinear discrete dynamic optimization problems.
• To present a direct method for assigning neural network weights based on established principles.
• To demonstrate the advantages of this neural network approach over conventional methods.

Main Methods:

• Development of a two-layer feedback artificial neural network with generalized recurrent neurons.
• A direct weight assignment method grounded in Bellmann's Optimality Principle and synaptic information exchange.
• Application of the neural network algorithm to dynamic programming problems.

Main Results:

• The proposed neural network effectively solves nonlinear discrete dynamic optimization problems.
• The direct weight assignment method provides a systematic approach to network configuration.
• Illustrative examples demonstrate successful application to shortest path and fuzzy decision-making problems.

Conclusions:

• The developed artificial neural network provides an advantageous approach for dynamic programming.
• The inherent parallelism of neural networks mitigates computational challenges.
• This method offers a promising alternative for complex optimization tasks.