Ancient I Ching sequence fails to improve neural network training

A new paper explores the statistical properties of the King Wen sequence, an ancient ordering of the I Ching hexagrams, to see if it could improve neural network training. Researchers found the sequence has distinct statistical characteristics, such as high transition distance and negative autocorrelation, which superficially resemble principles of curriculum learning. However, experiments across different hardware platforms and training methods showed that applying the King Wen sequence either degraded performance or had no significant effect, suggesting its unique variance destabilizes gradient-based optimization. AI

IMPACT Demonstrates that ancient ordering principles do not necessarily translate to improved AI training dynamics.

RANK_REASON The cluster contains a research paper detailing statistical analysis and experimental results related to machine learning. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• Apple Silicon
• Augustin Chantrel
• curiosity-driven exploration
• Curriculum learning
• I Ching
• King Wen sequence
• Mlx
• Monte Carlo permutation analysis
• neural network training
• NVIDIA RTX 2060
• PyTorch

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →