MODENN: A Shallow Broad Neural Network Model Based on Multi-Order Descartes Expansion

Area of Science:

• Artificial Intelligence
• Machine Learning
• Neural Networks

Background:

• Deep neural networks (DNNs) face limitations in parallelism, learning speed, performance, and reliability due to their hierarchical structure.
• The human brain's parallel processing inspires alternative neural network architectures.
• Addressing DNN weaknesses requires novel approaches for efficient feature extraction and classification.

Purpose of the Study:

• To propose a shallow, broad neural network model inspired by brain architecture.
• To introduce a novel feature extraction method using multi-order Descartes expansion.
• To demonstrate the effectiveness of the proposed model as an alternative to traditional DNNs.

Main Methods:

• A multi-order Descartes expansion operation is designed for efficient feature extraction.
• This operation transforms data into a high-dimensional feature space, enhancing pattern separability.
• A single-layer perceptron is combined with the Descartes expansion to create the Multi-Order Descartes Expansion Neural Network (MODENN).

Main Results:

• MODENN demonstrates capabilities equivalent to traditional multi-layer perceptrons and DNNs.
• Experimental results show MODENN's potential in implementability, parallelizability, performance, robustness, and interpretability.
• The model effectively handles classification tasks using enhanced feature separability.

Conclusions:

• MODENN presents a viable and potentially superior alternative to mainstream deep neural networks.
• The model's shallow and broad structure, combined with Descartes expansion, overcomes DNN limitations.
• MODENN offers significant advantages in parallel processing, learning efficiency, and model interpretability.