AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Decomposing Representation Space into Interpretable Subspaces with Unsupervised Learning

Xinting Huang, Michael Hahn

ICLR 2026 · Label-efficient and robust learning › Recognition and long-tail · approach: Contrastive representation theory · 3 citations (OpenAlex)

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
use NDM subspaces instead of an identity partitionGPT-2 Small subspace partition · GPT-2 Small test suite (average)Gini coefficient: 0.21 -> 0.71Improved
use NDM subspaces instead of a random orthogonal partitionGPT-2 Small subspace partition · GPT-2 Small test suite (average)Gini coefficient: 0.21 -> 0.71Improved
use NDM instead of PCA-based partition (PCA 1)GPT-2 Small subspace partition · GPT-2 Small test suite (average)Gini coefficient: 0.38 -> 0.71Improved
use NDM instead of clustering SAE featuresGPT-2 Small subspace partition · GPT-2 Small test suite (average)Gini coefficient: 0.56 -> 0.71Improved
use NDM instead of SAE feature clustering on test 4GPT-2 Small subspace partition · GPT-2 Small test suite, test 4Gini coefficient: 0.74 -> 0.50Hurt
use NDM instead of SAE feature clustering on test 3GPT-2 Small subspace partition · GPT-2 Small test suite, test 3Gini coefficient: 0.73 -> 0.63Hurt
use NDM instead of PCA 2 on test 1GPT-2 Small subspace partition · GPT-2 Small test suite, test 1Gini coefficient: 0.66 -> 0.89Improved
increase group sparsity of toy features from 0.25 to 0.75Toy superposition model · Toy superposition modelorthogonality between groupsHurt
lower hidden dimension from 12 to 10 in toy model with four groupsToy superposition model · Toy superposition model (group=(5,15,5,15))FVU: 0.066 -> 0.166Hurt
raise hidden dimension from 12 to 16 in toy model with four groupsToy superposition model · Toy superposition model (group=(5,15,5,15))FVU: 0.066 -> 0.016Improved
use Euclidean distance instead of cosine similarity in NDM objective (ablation)NDM · Toy modelpartition qualityImproved
search nearest neighbors among a small number of activations (N=4) in toy setting (ablation)NDM · Toy modelpartition qualityImproved

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