Run-off election-based decision method for the training and inference process in an artificial neural network.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Artificial neural networks (ANNs) classify unstructured data using activation functions and vector-matrix multiplication.
• Traditional sequence-based decision rules can misclassify similar data due to limited consideration of activation function configurations.

Purpose of the Study:

• To enhance the training and inference performance of ANNs.
• To mitigate classification errors in similar data by introducing a novel decision rule.

Main Methods:

• Implemented a run-off election-based decision rule with an additional filter evaluation.
• The filter evaluation was selected based on differences in common features of classified images.
• Applied the new algorithm to three types of shoe image datasets within a fully connected single-layer network.

Main Results:

• The novel decision rule achieved a recognition accuracy of approximately 82.03%.
• This performance surpasses the 79.23% accuracy obtained using the traditional sequence-based decision rule.
• The independent filter precisely supplied the output class in the decision step.

Conclusions:

• The run-off election-based decision rule with filter evaluation significantly improves ANN classification accuracy.
• This approach effectively addresses limitations of sequence-based rules in distinguishing similar data.
• The developed training algorithm offers precise output class determination in fully connected networks.