Research map / Label-efficient and robust learning / Noisy labels
Deep Learning From Crowdsourced Labels: Coupled Cross-Entropy Minimization, Identifiability, and Regularization
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 feature-geometry regularization to plain CCEM (ablation) | GeoCrowdNet(F) · LabelMe | test accuracy: 80.35 ± 0.40 -> 85.85 ± 0.33 | Helped |
| add annotator-geometry regularization to plain CCEM (ablation) | GeoCrowdNet(W) · Music | test accuracy: 65.93 ± 1.81 -> 66.46± 1.77 | Helped |
| use feature-output geometry regularization instead of annotator-matrix geometry regularization | GeoCrowdNet · LabelMe | test accuracy: 81.59 ± 0.73 -> 85.85 ± 0.33 | Helped |
| use feature-output geometry regularization instead of annotator-matrix geometry regularization | GeoCrowdNet · Music | test accuracy: 66.46± 1.77 -> 67.13± 1.27 | Helped |
| use feature-output geometry regularization rather than annotator-matrix geometry regularization | GeoCrowdNet · MNIST (5000,5) | test accuracy: 77.79 ± 8.97 -> 85.92 ± 2.73 | Helped |
| use feature-output geometry regularization rather than annotator-matrix geometry regularization | GeoCrowdNet · Fashion-MNIST (1000,30) | test accuracy: 85.56 ± 1.91 -> 84.47 ± 1.64 | Hurt |
| increase the number of labeled items from 1000 to 10000 with five annotators | GeoCrowdNet(F) · MNIST | test accuracy: 71.66 ± 4.48 -> 87.21 ± 2.47 | Helped |
| add feature-geometry regularization to plain CCEM (ablation) | GeoCrowdNet(F) · Music | test accuracy: 65.93 ± 1.81 -> 67.13± 1.27 | Helped |
Closest papers by meaning
- Coupled Confusion Correction: Learning from Crowds with Sparse Annotations (AAAI 2024 · similarity 0.87)
- Multi-Instance Multi-Label Classification from Crowdsourced Labels (AAAI 2025 · similarity 0.86)
- Unbiased Multi-Label Learning from Crowdsourced Annotations (ICML 2024 · similarity 0.85)
- IOFM: Using the Interpolation Technique on the Over-Fitted Models to Identify Clean-Annotated Samples (AAAI 2024 · similarity 0.84)
- Noisy Label Learning with Instance-Dependent Outliers: Identifiability via Crowd Wisdom (NeurIPS 2024 · similarity 0.84)
- Generative Cross-Entropy: A Strictly Proper Loss for Data-Efficient Classification (NeurIPS 2026 · 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.