Research map / Label-efficient and robust learning / Semi-supervised learning
A Data-Centric Approach for Improving Ambiguous Labels with Combined Semi-Supervised Classification and Clustering
Lars Schmarje, Monty Santarossa, Simon-Martin Schröder, Claudius Zelenka, Rainer Kiko, Jenny Stracke, Nina Volkmann, Reinhard Koch
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 |
|---|---|---|---|
| use argmax of l-hat as label instead of a sampled label | CE · Plankton | macro F1-Score: 86.71 -> 88.35 | Improved |
| use argmax of l-hat as label instead of a sampled label | FixMatch · Plankton | macro F1-Score: 80.29 -> 90.24 | Improved |
| use argmax of l-hat as label instead of a sampled label | CE · CIFAR-10H | macro F1-Score: 67.71 -> 68.89 | Improved |
| remove ambiguous images from the dataset | Mean-Teacher · Plankton | macro F1-Score: 88.94 -> 96.00 | Improved |
| remove ambiguous images from the dataset | FixMatch · CIFAR-10H | macro F1-Score: 79.15 -> 90.37 | Improved |
| remove ambiguous images from the dataset | CE · CIFAR-10H | macro F1-Score: 68.89 -> 86.57 | Improved |
| add DC3 to Mean-Teacher at 10% labeled | Mean-Teacher · MiceBone variant | best inner distance d: 0.4639 -> 0.2590 | Improved |
| add DC3 to Mean-Teacher at 100% labeled | Mean-Teacher · MiceBone variant | best F1: 0.7254 -> 0.5692 | Hurt |
| add DC3 to Mean-Teacher at 50% labeled | Mean-Teacher · MiceBone variant | best F1: 0.7348 -> 0.6862 | Hurt |
| add CE-1 loss to classification and clustering (CC) (ablation) | DC3 · pooled ablation (excluding FixMatch and Mice Bone) | clustering inner distance | Improved |
| add CE-1 loss to CC to form full DC3 (ablation) | CC only · pooled ablation (excluding FixMatch and Mice Bone) | stability, non-degenerated runs | Improved |
| use realistic prior ambiguity instead of p_A = 0.6 (ablation) | DC3 · pooled ablation (excluding FixMatch and Mice Bone) | F1-Score and inner distance d | Mixed |
Closest papers by meaning
- Is one annotation enough? - A data-centric image classification benchmark for noisy and ambiguous label estimation (NeurIPS 2022 · similarity 0.87)
- All Labels Are Not Created Equal: Enhancing Semi-Supervision via Label Grouping and Co-Training (CVPR 2021 · similarity 0.86)
- ClusMatch: Improving Deep Clustering by Unified Positive and Negative Pseudo-Label Learning (TPAMI 2025 · similarity 0.86)
- InterLUDE: Interactions between Labeled and Unlabeled Data to Enhance Semi-Supervised Learning (ICML 2024 · similarity 0.85)
- CGMatch: A Different Perspective of Semi-supervised Learning (CVPR 2025 · similarity 0.84)
- Label-Efficient Dataset Pruning via Semi-Supervised Pseudo-Labeling (NeurIPS 2026 · similarity 0.84)
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.
