Single dendritic neural classification with an effective spherical search-based whale learning algorithm.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Computational Neuroscience

Background:

• McCulloch-Pitts neuron models are standard in deep learning but are criticized for oversimplification.
• Dendritic Neuron Models (DNMs) offer advanced non-linear processing, improving prediction and classification.
• Backpropagation (BP) methods are prone to local minima and sensitive to initial conditions.

Purpose of the Study:

• To propose an innovative hybrid approach for co-evolving Dendritic Neuron Models (DNMs).
• To contrast this co-evolutionary method with traditional Backpropagation (BP) techniques.
• To enhance the computational power of DNMs for next-generation deep learning.

Main Methods:

• A hybrid approach co-evolves DNMs using an improved whale optimization algorithm with spherical search learning.
• Dynamic hybridizing is employed for co-evolution.
• Eleven classification datasets from the UCI Machine Learning Repository were used for validation.

Main Results:

• The proposed co-evolutionary method demonstrated superior classification accuracy compared to 10 non-BP methods and BP.
• Statistical analysis, including convergence speed, Wilcoxon sign-rank tests, ROC curves, and AUC, validated the model's efficiency.
• Well-learned DNMs were found to be computationally more potent than conventional McCulloch-Pitts neurons.

Conclusions:

• The hybrid co-evolutionary approach effectively overcomes limitations of BP methods.
• DNMs, when co-evolved, represent a significant advancement over McCulloch-Pitts neurons.
• This method offers a promising direction for developing next-generation deep learning models.