AI Paper Atlas

Research map / Label-efficient and robust learning / Semi-supervised learning

Language Semantic Graph Guided Data-Efficient Learning

Wenxuan Ma, Shuang Li, Lincan Cai, Jingxuan Kang

NeurIPS 2023 · Label-efficient and robust learning › Semi-supervised learning · approach: Self-training 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
add language semantic graph guided training (LSG) to vanilla fine-tuningResNet-50 fine-tuning · FGVC Aircraft 15% labelsaccuracy (%): 41.6 -> 55.6Improved
add regularization loss L_r (ablation)LSG · FGVC Aircraft 15% labelsaccuracy (%): 44.2 -> 55.6Improved
add alignment loss L_align (ablation)LSG · FGVC Aircraft 15% labelsaccuracy (%): 50.7 -> 55.6Improved
use trainable classifier instead of fixed language head (ablation)LSG · FGVC Aircraft 15% labelsaccuracy (%): 47.9 -> 55.6Improved
add stop gradient in L_r (ablation)LSG · FGVC Aircraft 15% labelsaccuracy (%): 48.3 -> 55.6Improved
replace GCN-based alignment with text prototype alignment (ablation)LSG · FGVC Aircraft 15% labelsaccuracy (%): 55.6 -> 51.1Hurt
replace GCN-based alignment with contrastive alignment (ablation)LSG · FGVC Aircraft 15% labelsaccuracy (%): 55.6 -> 54.2Hurt
refine label embeddings with GCN (ablation)label embeddings · FGVC AircraftCalinski-Harabasz index: 86.5 -> 1364.1Improved
substitute GAT for GCN (ablation)LSG · FGVC Aircraft SSLaccuracyMixed
vary edge-connection threshold tau (cross-label edge ratio) (ablation)LSG · image datasetsprimary model accuracyMixed
extend LSG with unlabeled data (semi-supervised)LSG · FGVC Aircraft 15% labelsaccuracy (%): 55.6 -> 71.4Improved
add LSG to Tip-Adapter style CLIP tuningCLIP ViT-B · FGVC Aircraft 15% labelsaccuracy (%): 36.9 -> 48.9Improved

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