Research map / Label-efficient and robust learning / Recognition and long-tail
GenKL: An Iterative Framework for Resolving Label Ambiguity and Label Non-conformity in Web Images Via a New Generalized KL Divergence
Xia Huang, Kai Fong Ernest Chong
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 |
|---|---|---|---|
| add iterative relabeling (multiple iterations) (ablation) | GenKL · Clothing1M | accuracy (%): 80.94 -> 81.34 | Improved |
| add double-hot vector non-normalization (keep vectors non-stochastic) (ablation) | GenKL · Clothing1M | accuracy (%): 81.18 -> 81.34 | Improved |
| add stratified sampling (ablation) | GenKL · Clothing1M | accuracy (%): 81.25 -> 81.34 | Improved |
| average prediction vectors over multiple models (ablation) | GenKL · Clothing1M | accuracy (%): 81.31 -> 81.34 | No clear effect |
| alpha 0.90 or 1.20 instead of 1.05 (ablation) | GenKL · Clothing1M | accuracy (%): 81.34 -> 81.20 (0.90), 81.19 (1.20) | Hurt |
| beta 0.02 or 0.04 instead of 0.03 (ablation) | GenKL · Clothing1M | accuracy (%): 81.34 -> 81.14 (0.02), 81.12 (0.04) | Hurt |
| increase |P| of uniform-like vectors from 1 to 20 (ablation) | GenKL · Clothing1M | accuracy (%): 80.26 -> 81.34 | Improved |
| sigma 0.05 instead of 0.01 (ablation) | GenKL · Clothing1M | accuracy (%): 81.14 -> 81.34 | Improved |
| increase omega2 from 1 to 32 (ablation) | GenKL · Clothing1M | accuracy (%): 80.26 -> 81.34 | Improved |
| use MSE loss instead of cross-entropy (ablation) | GenKL · Clothing1M | accuracy (%): 81.34 -> 80.69 | Hurt |
| use MAE loss instead of cross-entropy (ablation) | GenKL · Clothing1M | accuracy (%): 81.34 -> 80.16 | Hurt |
| use KL loss instead of cross-entropy (ablation) | GenKL · Clothing1M | accuracy (%): 81.34 -> 80.11 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 4 of 4 on most shared results.
- Jo-SRC — Jo-SRC: A Contrastive Approach for Combating Noisy Labels (CVPR 2021) · better on 4 of 4 shared results
- DivideMix — DivideMix: Learning with Noisy Labels as Semi-supervised Learning (ICLR 2020) · better on 4 of 4 shared results
- MetaCleaner — MetaCleaner: Learning to Hallucinate Clean Representations for Noisy-Labeled Visual Recognition (CVPR 2019) · better on 2 of 2 shared results
- Mopro — MoPro: Webly Supervised Learning with Momentum Prototypes (ICLR 2021) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Hyperspherical Learning in Multi-Label Classification (ECCV 2022 · similarity 0.86)
- Acknowledging the Unknown for Multi-Label Learning with Single Positive Labels (ECCV 2022 · similarity 0.85)
- Webly Supervised Image Classification with Self-Contained Confidence (ECCV 2020 · similarity 0.85)
- Heterogeneous Label Shift: Theory and Algorithm (ICML 2025 · similarity 0.83)
- Multi-Label Test-Time Adaptation with Bound Entropy Minimization (ICLR 2025 · similarity 0.83)
- Harnessing Hierarchical Label Distribution Variations in Test Agnostic Long-tail Recognition (ICML 2024 · similarity 0.83)
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.
