O(log2 M) self-organizing map algorithm without learning of neighborhood vectors

Area of Science:

• Computational science
• Bioinformatics
• Machine learning

Background:

• The original Self-Organizing Map (SOM) algorithm by Kohonen has a computational complexity of O(M^2).
• Traditional SOMs require neighborhood functions, adding computational overhead and parameter tuning complexity.

Discussion:

• A new SOM algorithm is introduced with a computational cost of O(log2 M).
• This algorithm integrates subdividing and binary search methods, eliminating the need for neighborhood functions.
• The removal of neighborhood functions reduces computational cost and simplifies parameter adjustment.

Key Insights:

• The proposed SOM algorithm achieves a significant reduction in computational complexity.
• Eliminating neighborhood functions streamlines the learning process and reduces parameter tuning.
• The algorithm's effectiveness is demonstrated through codon frequency analysis in Escherichia coli (E. coli) K12 genes.

Outlook:

• The reduced computational cost and simplified application of this new SOM algorithm can benefit diverse scientific fields.
• Further research can explore its application in other areas requiring efficient high-dimensional data analysis.
• This advancement has the potential to accelerate discoveries in computational biology and machine learning.