Layer-wise Cosine Similarity Analysis

Geometric analysis tracking how persona vector directions evolve across transformer layers to identify the transition layer

Neighborhood — ranked by edge-count

paper

method

Pairwise Cosine Similarity Analysis
related_to
Used to quantify the semantic clustering of adjective-set embeddings across model families and conditions

cosine ≥ 0.65 · no typed edge

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

Cosine Similarity Measurementmethod0.820
Used to measure alignment between DIM direction and cone basis vectors to assess overlap
Masked Cosine Similaritymethod0.792
Cosine similarity between feature activations restricted to tokens where one of the features fires; used to identify feature splitting relationships
Cosine Similarity Binary Classifiermethod0.791
Classifier using cosine similarity between activation vectors and steering vectors to detect deception with 89% accuracy
Cosine Similarity-Based Deception Detectionconcept0.790
Detection mechanism computing cosine similarity between activation vectors and steering vectors to classify deception
Feature neighborhood exploration via cosine similarity of decoder weightsmethod0.781
Identifying related features by cosine distance in SAE decoder space.
Cosine Similarity Ranking for Instruction Discoverymethod0.779
Method to discover new reflection-inducing instructions by ranking candidate tokens by cosine similarity to steering vectors.
layer-wise trajectory analysismethod0.774
Computing per-layer S(ℓ) to summarize geometry.
Cosine similarity between truth probesmethod0.772
Geometric evaluation of truth direction alignment across layers and prompt templates.