Adaptive learning algorithm and its convergence analysis with complex-valued error loss network.

Area of Science:

• Machine Learning
• Supervised Learning
• Signal Processing

Background:

• Loss functions are critical for evaluating machine learning model performance.
• Existing models face challenges with complex-valued signals and parameters.
• Adaptive learning algorithms require stable convergence properties.

Purpose of the Study:

• Introduce a novel Complex Error Loss Network (CELN) for supervised learning.
• Address limitations in handling complex-valued data.
• Investigate the convergence properties of the CELN adaptive learning algorithm.

Main Methods:

• Developed a novel Complex Error Loss Network (CELN).
• Applied the contraction mapping theorem to analyze algorithm convergence.
• Evaluated CELN performance against benchmark methods.

Main Results:

• CELN reduces prediction error by at least 4.1% compared to benchmarks.
• The adaptive learning algorithm demonstrates stable convergence towards optimal solutions.
• CELN maintains performance stability in non-Gaussian noise environments.

Conclusions:

• CELN offers a robust solution for supervised learning with complex-valued data.
• The model's convergence properties ensure reliable optimization.
• CELN presents a significant advancement in machine learning loss function design.