Cycle k-NN

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

• LCS k-NNmethod0.790
  Calculates the longest common subsequence of nearest neighbors normalized by sequence length
• Edit k-NNmethod0.789
  Computes edit distance required to match nearest neighbors between two datasets, normalized by maximum edit distance
• Longest Common Subsequence k-NNmethod0.762
  Alternative alignment metric compared in appendix; calculates longest common subsequence of nearest neighbor lists
• Edit Distance k-NNmethod0.744
  Alternative alignment metric compared in appendix; computes edit distance between nearest neighbor lists
• K-Compositionconcept0.706
  A form of attention head composition where W_K reads from a subspace affected by a previous head; central to how induction heads are implemented
• Four-step cycle (context, latent centers, possible action, new construction)method0.687
  An iterated design process: 1) observe current configuration, 2) identify latent centers, 3) decide where to build to strengthen a latent center, 4) construct, take the whole to a new plateau.
• Two-stage cyclic computationconcept0.684
  Llama's discovered algorithm: first compute integer sum via base-10 addition, then map the sum back to cyclic concept space
• Cyclic Geometryconcept0.675
  One of the geometric structures used to characterize reasoning task representations in language models, used as a test case for manifold steering