framework
active
framework:arcface-cosface-angular-margin-lossArcFace/CosFace Angular Margin Loss
Angular margin technique borrowed from face recognition and applied to the prototype contrast loss in SAE training
Neighborhood — ranked by edge-count
Papers (1)
paper
Methods (1)
method
- Prototype Contrast Loss (L_CE)implementsLoss function pulling representations toward positive centroid and pushing away from negative centroid with angular margins
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- The outer edge space that protects the text block but can be invaded by marginalia.
- Explicitly identified limitation of the proposed mitigation method
- Auxiliary objective combining L2 and cosine losses against pre-recorded CL vectors to improve causal relevance when one model is causally inaccessible.
- Addressing disparity in loss magnitudes across tasks at the loss level
- Feature extraction method computing cosine similarity of hidden representations with reflection direction across all layers
- Auxiliary training objective from Grant (2025) that constrains intervened representations to remain near natural distribution
- Margins function actively, not neutrally; they define boundaries and create the distinction between text-space and world-space.
- Cosine similarity between feature activations restricted to tokens where one of the features fires; used to identify feature splitting relationships