Research map / Label-efficient and robust learning / Recognition and long-tail
Weakly Supervised Collective Feature Learning From Curated Media
Yusuke Mukuta, Akisato Kimura, David Adrian, Zoubin Ghahramani
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 changed | Where | Measured | Effect |
|---|---|---|---|
| Fuse VGG16 with fine-tuned curated-feature representations | VGG16 features · Pinterest, UECFOOD100, UECFOOD256, Apparel, Hipster, Instagram | MAP or ACC: 0.430 -> 0.477; 0.511 -> 0.536; 0.468 -> 0.502; 0.576 -> 0.586; 0.592 -> 0.642; 0.910 -> 0.919 | Improved |
| Add sparse-coding curated-group features to VGG16 | VGG16 features · Pinterest, UECFOOD100, UECFOOD256, Apparel, Hipster, Instagram | MAP or ACC: 0.430 -> 0.445; 0.511 -> 0.514; 0.468 -> 0.473; 0.576 -> 0.566; 0.592 -> 0.599; 0.910 -> 0.916 | Mixed |
| Use sparse-coding curated-group features instead of VGG16 features alone | VGG16 · Pinterest, UECFOOD100, UECFOOD256, Apparel, Hipster | MAP or ACC: 0.430 -> 0.409; 0.511 -> 0.454; 0.468 -> 0.376; 0.576 -> 0.488; 0.592 -> 0.552 | Hurt |
| Fine-tune VGG16 using curated-group link prediction | VGG16 · Hipster | ACC: 0.592 -> 0.615 | Improved |
| Fine-tune VGG16 using curated-group link prediction | VGG16 · Pinterest, UECFOOD100, UECFOOD256, Apparel, Instagram | MAP or ACC: 0.430 -> 0.404; 0.511 -> 0.453; 0.468 -> 0.398; 0.576 -> 0.470; 0.910 -> 0.896 | Hurt |
| Add curated-group latent features to link prediction (ablation) | Link prediction model · Small and large link-prediction graphs | mean AUC: 0.895 -> 0.928; 0.8442 -> 0.8482 | Improved |
| Add image latent features to link prediction (ablation) | Link prediction model · Small and large link-prediction graphs | mean AUC: 0.895 -> 0.896; 0.8442 -> 0.8460 | No clear effect |
| Add both image and curated-group latent features (ablation) | Link prediction model · Small and large link-prediction graphs | mean AUC: 0.895 -> 0.929; 0.8442 -> 0.8477 | Improved |
| Replace the baseline link-prediction model with the proposed model without latent features (ablation) | Link prediction model · Small link-prediction graph | mean AUC: 0.895 -> 0.898 | Improved |
| Replace the baseline link-prediction model with the proposed model without latent features (ablation) | Link prediction model · Large link-prediction graph | mean AUC: 0.8442 -> 0.8441 | No clear effect |
Closest papers by meaning
- Learning Weakly-supervised Contrastive Representations (ICLR 2022 · similarity 0.85)
- Robust Detection of Link Communities in Large Social Networks by Exploiting Link Semantics (AAAI 2018 · similarity 0.84)
- Weakly-Supervised Deep Nonnegative Low-Rank Model for Social Image Tag Refinement and Assignment (AAAI 2017 · similarity 0.84)
- Learning Visual Features from Large Weakly Supervised Data (ECCV 2016 · similarity 0.84)
- A Comprehensive Study on Text-attributed Graphs: Benchmarking and Rethinking (NeurIPS 2023 · similarity 0.84)
- SSTAG: Structure-Aware Self-Supervised Learning Method for Text-Attributed Graphs (NeurIPS 2025 · similarity 0.82)
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.
