DAS behavioral loss produces EMD along feature dimensions of 0.032±0.003 on synthetic 10-class dataset

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

• DAS behavioral loss achieves IIA of 0.997±0.001 on synthetic 10-class dataset training/test setsfinding0.854
  IIA baseline for DAS behavioral loss on synthetic dataset
• For small CL loss weights epsilon, IIA is maintained (potentially improved) while EMD decreases in Boundless DAS on a 7B LLMfinding0.766
  Empirical result showing the CL loss can reduce divergence without sacrificing interpretability accuracy
• Modified CL loss outperforms behavioral DAS loss in OOD transfer from dense to sparse class partitionfinding0.761
  Key practical utility result: CL loss improves generalization of alignment to out-of-distribution settings
• Representational divergence (as measured by EMD) can predict lower out-of-distribution intervention performanceclaim0.755
  Practical utility of reducing divergence demonstrated through regression analysis
• Boundless DAS interchange interventions produce EMD exceeding natural-natural baselinefinding0.748
  Empirical demonstration that DAS interventions produce divergent representations
• Linear regression of OOD IIA on training EMD yields coefficient -0.3424, R^2=0.729, F(1,28)=75.28, p<.001finding0.745
  Statistical evidence that training divergence (EMD) predicts lower OOD intervention performance
• GRU behavior can be compressed to as few as 4 dimensions using DAS and MAS with comparable IIAsfinding0.742
  Shows that behaviorally relevant information is low-dimensional; contrasted with model stitching achieving near-perfect IIA at rank 2.
• Mean difference patching on Llama-3-8B layer 10 produces intervened EMD exceeding the natural-natural baselinefinding0.726
  Empirical demonstration that MDVP produces divergent representations in a real LLM