AI Paper Atlas

Research map / Label-efficient and robust learning / Self-supervised and contrastive

You Never Cluster Alone

Yuming Shen, Ziyi Shen, Menghan Wang, Jie Qin, Philip Torr, Ling Shao

NeurIPS 2021 · Label-efficient and robust learning › Self-supervised and contrastive · approach: Contrastive versus non-contrastive theory · 1 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
remove cluster-level contrastive loss L1 (ablation)TCC · CIFAR-10ACC: 90.6 -> 78.7Hurt
remove instance-level loss L2, keep only L1 (ablation)TCC · CIFAR-10ACC: 90.6 -> 45.4Hurt
sample 10 Gumbel softmax latents instead of one (ablation)TCC · CIFAR-10ACC: 90.6 -> 90.1No clear effect
remove cluster-level memory bank P and use K-1 other clusters as negatives (ablation)TCC · CIFAR-10ACC: 90.6 -> 82.9Hurt
add cluster-level augmentation for L1 (ablation)TCC · CIFAR-10ACC: 85.3 -> 90.6Improved
use soft assignments rather than hard assignments for cluster aggregation (ablation)TCC · CIFAR-10ACC: 79.2 -> 90.6Improved
use end-to-end batch training instead of alternating epoch-wise training (ablation)TCC · CIFAR-10ACC: 80.0 -> 90.6Improved
add cluster-level representations over MiCE instance-only modelTCC · CIFAR-10ACC: 83.5 -> 90.6Improved
TCC versus CCTCC · STL-10ACC: 85.0 -> 81.4Hurt
TCC versus CCTCC · Tiny ImageNetACC: 14.0 -> 30.6Improved
TCC versus SwAV re-implemented with same backboneTCC · CIFAR-10ACC: 79.6 -> 90.6Improved
larger loss weight alpha for L1TCC · CIFAR-10ACCNo clear effect

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