Kendall's tau rank correlation

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

• Rank-Order Correlation (Kendall's rho)method0.821
  Statistical method used by Yodan Rose to measure agreement between different people's neighborhood diagnoses.
• Spearman's Rank Correlationmethod0.780
  Statistical measure used to evaluate correlation between diversity metrics and diversity parameter / human judgments
• Spearman Rank Correlationmethod0.776
  Used to compare RDMs in RSA computations; noted to have sensitivity issues with differing relative extrema in embedding layers.
• Activation Correlationmethod0.704
  Pearson correlation of feature activations across 40M tokens used to measure feature similarity and universality across models
• Spearman's rank correlation among different alignment metrics (CKA, SVCCA, Mutual k-NN, CKNNA) over 78 vision models is high across variants, with all p-values below 2.24×10^-105finding0.702
  Validates robustness of alignment metric choice
• Token-100 correlation persistence metricmethod0.692
  Measures emotion feature persistence as correlation between z-scored activation at token 0 and token 100 across all eligible target model tokens
• Kendall's τ = 0.76 (p<.001) for Conscientiousness dimension LLM scoring vs human judgmentfinding0.687
  Validates GPT-4o scoring reliability for Conscientiousness personality dimension
• Most correlated neuron A/neurons/470 has correlation of only 0.18 with base64 feature A/1/2357 and responds to code, HTML labels, URLsfinding0.676
  Shows base64 feature is polysemantic at neuron level but monosemantic as learned feature