method
active
method:locally-masked-convolutional-crosscoderLocally masked convolutional crosscoder
Each feature assigned a window of K layers to reduce training FLOPs.
Neighborhood — ranked by edge-count
Methods (1)
method
- Sparse CrosscoderextendsSparse dictionary learning variant that reads and writes to multiple layers or models, yielding shared features.
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.
- Learn a shared feature space across two models; basis of the work.
- Self-supervised learning method that optimizes reconstruction tasks; included in the paper's analysis as a multi-task objective
- MLP replacement with sparse features underlying the original attribution graphs.
- Mask virtual weights during training targeting interference.
- Basic crosscoder reading and reconstructing all layers jointly.
- Baseline from Mishra-Sharma et al. dedicating features to shared variance.
- Train a single SAE encoding and decoding activations from two models simultaneously to find shared and exclusive features.
- Extension of SAEs that jointly learns latents across representations from different models; proposed as alternative for extended fine-tuning