AI Paper Atlas

Research map / Learning theory and algorithms / Clustering

Unsupervised Manifold Linearizing and Clustering

Tianjiao Ding, Shengbang Tong, Kwan Ho Ryan Chan, Xili Dai, Yi Ma, Benjamin D. Haeffele

ICCV 2023 · Learning theory and algorithms › Clustering · approach: Subspace clustering · 11 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
replace self-supervised feature initialization with random initialization (ablation)MLC · CIFAR-10clustering ACC (%): 86.3 -> 20.0Hurt
add MLC objective update after self-supervised initialization (vs EnSC on initialized features) (ablation)MLC · CIFAR-10clustering ACC (%): 73.4 -> 86.3Improved
add data augmentation when updating the MLC objective (ablation)MLC · CIFAR-10clustering ACC (%): 80.0 -> 86.3Improved
double batch size or number of augmentations (ablation)MLC · CIFAR-10clustering ACC: <= 1% improvementNo clear effect
run subspace clustering (EnSC) on MLC-learned features instead of TCR features (ablation)EnSC · CIFAR-10clustering ACC (%): 72.2 -> 81.5Improved
deterministic doubly stochastic membership initialization instead of random cluster head (NMCE) (ablation)MLC · CIFAR-10 over 10 seedsACC std: 3.69 -> 0.24Improved
MoCoV2 pre-training instead of TCR for MLC initialization (ablation)MLC · CIFAR-20clustering ACC: .522 -> .583Improved
use MLC instead of SCAN and IMC-SwAV on imbalanced clustersclustering · Imb-CIFAR-10clustering ACC (%): 65.7 -> 80.0Improved
MLC-TCR vs best baseline on CIFAR-10MLC · CIFAR-10clustering ACC: IMC-SwAV .891 vs MLC-TCR .863Hurt
use MLC-TCR on CIFAR-100 vs IMC-SwAV and SCANclustering · CIFAR-100ACC and runtime: ACC 43.9 -> 49.4; time 529.4 -> 284.4 minImproved

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