Research map / Label-efficient and robust learning / Noisy labels
Dual-Granularity Learning for Regression with Continuous Noisy Labels
Hui GUO, Boyu Wang, Grace Yi
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes a dual-granularity learning approach for regression with continuous noisy labels, addressing heterogeneous label-noise magnitudes across instances.
Problem
Learning regression models when continuous-valued training labels contain noise whose magnitude varies across instances, a setting less studied than noisy-label classification.
Approach
Uses two granularities of learning to reconcile conflicting needs: sensitivity to label values for correcting small-to-moderate deviations and robustness for large deviations (details not stated in the abstract excerpt).
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use GRAIN dual-granularity training instead of vanilla MSE | ResNet-50 · UTKFace std_max=60 | MAE: 9.94 -> 7.54 | Improved |
| use GRAIN instead of vanilla MSE under real noise | DenseNet-121 · Bone-Age | MAE: 13.56 -> 12.85 | Improved |
| add a histogram (BIN) head to baselines | baseline methods · UTKFace std_max=60/90 | MAE: Vanilla 9.94 -> 8.44 (std60); Selfie 8.30 -> 8.44; CoTeaching 7.98 -> 8.29 | Mixed |
| stronger data augmentation (GRAIN-AUG) | GRAIN · UTKFace std_max=30 | MAE: 7.01 -> 6.81 | Improved |
| noise-robust classification methods (Selfie, CNLCU, CoTeaching) adapted to regression | baselines · Bone-Age | MAE: Vanilla 13.56 vs Selfie 15.43, CNLCU-S 15.97 | Hurt |
| increase label noise from std_max 30 to 90 | Vanilla · UTKFace | MAE: 8.13 -> 10.25 | Hurt |
| ablate residual head, coarse contrastive term, Huber loss, pseudo-label variants (ablation) | GRAIN · UTKFace | MAE/RMSE/R2 | Improved |
Closest papers by meaning
- Learning from Noisy Labels via Conditional Distributionally Robust Optimization (NeurIPS 2024 · similarity 0.87)
- Understanding Domain Generalization: A Noise Robustness Perspective (ICLR 2024 · similarity 0.87)
- Gradient Aligned Regression via Pairwise Losses (ICML 2025 · similarity 0.86)
- Label Noise: Ignorance Is Bliss (NeurIPS 2024 · similarity 0.86)
- Robust Logit Adjustment for Learning with Long-Tailed Noisy Data (AAAI 2025 · similarity 0.86)
- Sample Selection via Contrastive Fragmentation for Noisy Label Regression (NeurIPS 2024 · similarity 0.86)
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.
