数据清洗论文趋势 2023–2026

主要结论

• 数据清洗论文从 2023 年的 21 篇降到 2026 年的 10 篇，热度在回落。
• ICML 是发数据清洗论文最多的会议，2023–2026 年共 17 篇，占四个会议总量的 35%。
• 在「以数据为中心 AI」的 5 个主题里，它的 2026 年论文数排第 5；同组增长最快的合成数据生成是 1.3 倍，数据清洗是 0.5 倍。

代表论文

按年份和会议均匀挑选，方便快速看到每一届的样貌

• NeurIPS 2026GradTrack: Detecting Noisy Labels via Temporal Trajectories of Class-wise Gradient Misalignment
• ICML 2026Learning Locally, Revising Globally: Global Reviser for Federated Learning with Noisy Labels
• ICLR 2026Enhancing Learning with Noisy Labels via Rockafellian Relaxation
• CVPR 2026Cleaning the Pool Progressive Filtering of Unlabeled Pools in Deep
• NeurIPS 2025Enhancing Sample Selection Against Label Noise by Cutting Mislabeled Easy Examples
• ICML 2025Free Process Rewards without Process Labels
• ICLR 2025Learning to Generate Diverse Pedestrian Movements from Web Videos with Noisy Labels
• CVPR 2025Directional Label Diffusion Model for Learning from Noisy Labels
• ICML 2024Learning with Complementary Labels Revisited: The Selected-Completely-at-Random Setting Is More Practical
• CVPR 2024Estimating Noisy Class Posterior with Part-level Labels for Noisy Label Learning
• NeurIPS 2023Label Correction of Crowdsourced Noisy Annotations with an Instance-Dependent Noise Transition Model
• ICML 2023Data Structures for Density Estimation
• ICLR 2023Learning to Segment from Noisy Annotations: A Spatial Correction Approach
• CVPR 2023Fine-Grained Classification With Noisy Labels
• NeurIPS 2026Memorization Is Folding: Topological Signatures of Noisy-Label Learning
• CVPR 2026Debiased Sample Selection for Learning with Noisy Labels
• NeurIPS 2025ELDET: Early-Learning Distillation with Noisy Labels for Object Detection
• ICML 2025Retraining with Predicted Hard Labels Provably Increases Model Accuracy
• ICML 2024Pi-DUAL: Using privileged information to distinguish clean from noisy labels
• CVPR 2024Learning with Structural Labels for Learning with Noisy Labels

全部数据清洗论文

共 17 篇，第 1 / 1 页

1. ICML 2026Learning Locally, Revising Globally: Global Reviser for Federated Learning with Noisy Labels
2. ICML 2025Free Process Rewards without Process Labels
3. ICML 2025Retraining with Predicted Hard Labels Provably Increases Model Accuracy
4. ICML 2024Consistent Diffusion Meets Tweedie: Training Exact Ambient Diffusion Models with Noisy Data
5. ICML 2024Learning with Complementary Labels Revisited: The Selected-Completely-at-Random Setting Is More Practical
6. ICML 2024Pi-DUAL: Using privileged information to distinguish clean from noisy labels
7. ICML 2024Provably Robust DPO: Aligning Language Models with Noisy Feedback
8. ICML 2024Self-cognitive Denoising in the Presence of Multiple Noisy Label Sources
9. ICML 20232D-Shapley: A Framework for Fragmented Data Valuation
10. ICML 2023Data Structures for Density Estimation
11. ICML 2023Data-Derived Weak Universal Consistency
12. ICML 2023Generating Private Synthetic Data with Genetic Algorithms
13. ICML 2023Gradient-Free Structured Pruning with Unlabeled Data
14. ICML 2023Learning Functional Distributions with Private Labels
15. ICML 2023Learning Noisy OR Bayesian Networks with Max-Product Belief Propagation
16. ICML 2023Mitigating Memorization of Noisy Labels by Clipping the Model Prediction
17. ICML 2023Unearthing InSights into Mars: Unsupervised Source Separation with Limited Data

常见问题

2023–2026 年四大会议一共有多少篇数据清洗论文？

共 48 篇，按年份分别是 2023 年 21 篇、2024 年 11 篇、2025 年 6 篇、2026 年 10 篇。

NeurIPS 2026 有多少篇数据清洗论文？

NeurIPS 2026 有 6 篇数据清洗论文，上一届是 2 篇。

ICLR 2026 有多少篇数据清洗论文？

ICLR 2026 有 1 篇数据清洗论文，上一届是 1 篇。

ICML 2026 有多少篇数据清洗论文？

ICML 2026 有 1 篇数据清洗论文，上一届是 2 篇。

CVPR 2026 有多少篇数据清洗论文？

CVPR 2026 有 2 篇数据清洗论文，上一届是 1 篇。

哪个会议的数据清洗论文最多？

ICML，2023–2026 年共 17 篇。