Multiorder neurons for evolutionary higher-order clustering and growth

Area of Science:

• Artificial Intelligence
• Machine Learning
• Computational Neuroscience

Background:

• Traditional clustering methods struggle with irregularly shaped data.
• Higher-order neurons use tensors to model complex neuron shapes, but have limitations.
• Existing methods lack adaptability in selecting optimal neuron orders for diverse datasets.

Discussion:

• Multiorder neurons extend higher-order neurons by using evolutionary algorithms to dynamically select the optimal neuron order.
• This approach addresses limitations of higher-order neurons by improving data distribution analysis.
• The method identifies the correct neuron order for specific pattern clusters, enhancing interpretability.

Key Insights:

• Evolutionary multiorder neurons significantly outperform self-organizing maps and higher-order neurons in clustering accuracy.
• Empirical results on Iris, Wine, and Glass datasets demonstrate the method's superior performance.
• The proposed method achieves higher correlation between found clusters and ground truth information.

Outlook:

• The development of an intuitive model for growing multiorder neurons to automatically determine the number of clusters.
• Potential applications in complex pattern recognition and data mining tasks.
• Further research into the theoretical underpinnings and broader applicability of evolutionary multiorder neurons.