An Extension Network of Dendritic Neurons.

Area of Science:

• Computational Neuroscience
• Machine Learning
• Artificial Intelligence

Background:

• Deep learning (DL) excels but faces challenges with model complexity and interpretability due to simplified neuron units.
• Biologically plausible dendritic neuron models (DNMs) offer improvements but are limited to binary classification tasks.

Purpose of the Study:

• To develop an extended dendritic neuron network (DNN) capable of solving multiple-class classification problems.
• To introduce an efficient error-back-propagation learning algorithm for the proposed network.
• To evaluate the proposed method's effectiveness and superiority against state-of-the-art classifiers.

Main Methods:

• Proposed a novel extended network architecture based on dendritic structures.
• Derived an efficient error-back-propagation learning algorithm for multi-class classification.
• Conducted extensive experiments on ten datasets, including a real-world web service quality application.

Main Results:

• The proposed extended dendritic neuron network demonstrated effectiveness and superiority compared to nine other state-of-the-art classifiers.
• The learning algorithm proved competent and reliable in classification performance and stability.
• The method showed a notable advantage in handling small-scale imbalanced datasets.

Conclusions:

• The novel extended dendritic neuron network effectively addresses multi-class classification challenges.
• The derived learning algorithm is robust, reliable, and advantageous for imbalanced data scenarios.
• The study validates the potential of biologically inspired neural network models in advanced machine learning tasks.