A training algorithm with selectable search direction for complex-valued feedforward neural networks.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Neural Networks

Background:

• Complex-valued feedforward neural networks (CVFFNNs) are increasingly used in signal processing and pattern recognition.
• Existing training algorithms may lack efficiency and flexibility in optimizing CVFFNNs.
• The need for advanced training methods is crucial for improving performance in complex data domains.

Purpose of the Study:

• To present an efficient training algorithm for complex-valued feedforward neural networks.
• To introduce a novel approach utilizing a tree structure and direction factors for enhanced optimization.
• To demonstrate the algorithm's effectiveness in improving convergence speed and solution accuracy.

Main Methods:

• The proposed algorithm employs a tree structure to manage the search space.
• Direction factors are introduced to enable flexible selection of search directions at each iteration.
• The algorithm iteratively reduces the objective function by selecting optimal search directions.

Main Results:

• The algorithm demonstrates faster convergence compared to well-known training methods.
• More accurate solutions are obtained due to the flexible determination of search directions.
• Experimental simulations validate the algorithm's effectiveness in pattern recognition, channel equalization, and complex function approximation.

Conclusions:

• The proposed tree-structure-based training algorithm offers an efficient and flexible method for optimizing complex-valued neural networks.
• The use of direction factors significantly enhances convergence speed and solution accuracy.
• The algorithm shows broad applicability in various complex-valued data processing tasks.