A Dendritic Neuron Model with Adaptive Synapses Trained by Differential Evolution Algorithm.

Area of Science:

• Computational Neuroscience
• Artificial Intelligence
• Hardware Implementation

Background:

• Dendritic neuron models (DNMs) offer a biologically plausible approach to computation.
• Adaptive synapses are crucial for learning and information processing in neural networks.
• Existing training algorithms may not fully exploit the potential of DNMs for complex tasks.

Purpose of the Study:

• To propose a dendritic neuron model with adaptive synapses (DMAS) trained using the differential evolution (DE) algorithm.
• To evaluate the performance of the DE-trained DMAS against established models.
• To demonstrate the conversion of the trained DMAS into a hardware-implementable logic circuit.

Main Methods:

• Development of a dendritic neuron model with adaptive synapses (DMAS).
• Training the DMAS using the differential evolution (DE) algorithm.
• Conversion of the trained DMAS into a logic circuit using basic logical devices.
• Experimental validation on five UCI Machine Learning Repository classification datasets.

Main Results:

• The DE-DMAS demonstrated superior performance across all evaluated metrics, including correct rate and convergence rate.
• Comparative analysis showed the DE-DMAS outperformed the backpropagation algorithm-trained dendritic neuron model (BP-DNM) and neural network (BPNN).
• The trained model's logic circuit implementation proved effective for solving complex nonlinear problems.

Conclusions:

• The differential evolution algorithm is highly effective for training dendritic neuron models with adaptive synapses.
• The proposed DE-DMAS offers a robust and efficient approach for complex nonlinear problem-solving.
• The model's conversion to logic circuits facilitates practical hardware implementation.