AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

SVD as a Fast Interpretability Method for Transformers

Min Xue, Artur Andrzejak

ICML 2026 · Architectures and efficiency › RNNs and normalization · approach: Sparse dictionary representations

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
reconstruct W_down with top-20 SCA subspaces only (sufficiency)NaNA · Dgene (GPT-2 family, middle-to-late layers)top-1 accuracy: 97-99%Improved
ablate top-k high-contribution subspaces (ablation)NaNA · Dgene (GPT-2 family)top-1 accuracy, probability, rankHurt
ablate random DEU subspaces instead of top-k (ablation)NaNA · GPT2-Medium positive semantics casetarget rankNo clear effect
remove top 10 subspaces (ablation)NaNA · GPT2-Medium 'The cat looks very'top predictionsHurt
reconstruct with only top-5 subspaces (ablation)NaNA · GPT2-Medium 'The cat looks very'target token recoveryImproved
amplify subspace 7 of layer 16 (alpha from 0 to 8) (ablation)NaNA · GPT2-Medium 'The cat looks very'top tokensImproved
suppress subspace 7 (alpha below 0) (ablation)NaNA · GPT2-Medium 'The cat looks very'top tokensHurt
use SCA subspaces instead of SAE and TransCoder features beyond k>60NaNA · GPT-2 layer 10target rankImproved
NaNA instead of SAE/TransCoder for single matrix runtimeNaNA · GPT-2 single projection matrixruntime: ≈30 min -> < 2 minImproved
NaNA instead of SAE/TransCoder GPU memoryNaNA · GPT-2 single projection matrixGPU memory: ≈16 GB -> < 1 GBImproved
NaNA instead of precomputing SAE/TransCoder modules for GPT2-XLNaNA · GPT2-XL pathway extractionruntime: over 24 hours -> 8.59 minutesImproved
add more subspaces to reconstruction (around k=30) (ablation)NaNA · GPT-2 layer 10target probabilityMixed

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy