弱监督论文趋势 2023–2026

主要结论

• 弱监督论文从 2023 年的 34 篇增长到 2026 年的 45 篇，三年间扩大到约 1.3 倍。
• NeurIPS 是发弱监督论文最多的会议，2023–2026 年共 42 篇，占四个会议总量的 31%。
• 交叉方向里涨得最快的是对齐与安全：同时属于两个主题的论文从 2023 年的 0 篇增至 2026 年的 7 篇。
• 在「以数据为中心 AI」的 5 个主题里，它的 2026 年论文数排第 2；同组增长最快的合成数据生成是 1.3 倍，弱监督是 1.3 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 2026Affine-invariant Cubic Newton with Weak Learners
• ICML 2026Weak-to-Strong Generalization via Bregman Bias–Variance Decomposition
• ICLR 2026Weak-to-Strong Generalization with Failure Trajectories
• CVPR 2026GEM-TFL Bridging Weak and Full Supervision for Forgery Localization through
• NeurIPS 2025Disentangled Cross-Modal Representation Learning with Enhanced Mutual Supervision
• ICML 2025How does Labeling Error Impact Contrastive Learning? A Perspective from Data Dimensionality Reduction
• ICLR 2025LASER: A Neuro-Symbolic Framework for Learning Spatio-Temporal Scene Graphs with Weak Supervision
• CVPR 2025Bridge the Gap: From Weak to Full Supervision for Temporal Action Localization with PseudoFormer
• NeurIPS 2024Metrizing Weak Convergence with Maximum Mean Discrepancies
• ICML 2024A General Framework for Learning from Weak Supervision
• ICLR 2024Tractable Probabilistic Graph Representation Learning with Graph-Induced Sum-Product Networks
• CVPR 2024Molecular Data Programming: Towards Molecule Pseudo-labeling with Systematic Weak Supervision
• NeurIPS 2023Latent Graph Inference with Limited Supervision
• ICML 2023AdaBoost is not an Optimal Weak to Strong Learner
• ICLR 2023Generative Modeling Helps Weak Supervision (and Vice Versa)
• CVPR 2023DPF: Learning Dense Prediction Fields With Weak Supervision
• NeurIPS 2026Clustering with Weak Distance Oracles
• ICML 2026Learning Permutation from Structure Without Supervision
• ICLR 2026Language-guided Open-world Video Anomaly Detection under Weak Supervision
• CVPR 2026Generalizable Video Quality Assessment via Weak-to-Strong Learning

全部弱监督论文

共 136 篇，第 1 / 3 页

1. CVPR 2026Beyond Weak Supervision MLLMs-Guided Graded Knowledge Distillation for Unsupervised Camouflaged
2. CVPR 2026Conversational Image Segmentation Grounding Abstract Concepts with Scalable Supervision
3. CVPR 2026FlexAvatar Learning Complete 3D Head Avatars with Partial Supervision
4. CVPR 2026GEM-TFL Bridging Weak and Full Supervision for Forgery Localization through
5. CVPR 2026Generalizable Video Quality Assessment via Weak-to-Strong Learning
6. CVPR 2026Improving Diffusion Generalization with Weak-to-Strong Segmented Guidance
7. CVPR 2026MatchED Crisp Edge Detection Using End-to-End Matching-based Supervision
8. CVPR 2026NexusFlow Unifying Disparate Tasks under Partial Supervision via Invertible Flow
9. CVPR 2026PropFly Learning to Propagate via On-the-Fly Supervision from Pre-trained Video
10. CVPR 2026UnReflectAnything RGB-Only Highlight Removal by Rendering Synthetic Specular Supervision
11. ICLR 2026Demystifying Supervision Data Generalization in Multimodal LMs
12. ICLR 2026Does Weak-to-strong Generalization Happen under Spurious Correlations?
13. ICLR 2026Language-guided Open-world Video Anomaly Detection under Weak Supervision
14. ICLR 2026Metric $k$-clustering using only Weak Comparison Oracles
15. ICLR 2026Scalable Oversight via Partitioned Human Supervision
16. ICLR 2026Truthfulness Despite Weak Supervision: Evaluating and Training LLMs Using Peer Prediction
17. ICLR 2026Weak-to-Strong Generalization with Failure Trajectories
18. ICLR 2026When Weak LLMs Speak with Confidence, Preference Alignment Gets Stronger
19. ICML 2026Compute as Teacher: Turning Inference Compute Into Reference-Free Supervision
20. ICML 2026DC-W2S: Dual-Consensus Weak-to-Strong Training for Reliable Process Reward Modeling in Biological Reasoning
21. ICML 2026Deep Trajectory Supervision: Deep Supervision Strikes Back
22. ICML 2026Learning Permutation from Structure Without Supervision
23. ICML 2026Preference-based Antibody Expression Ranking: Scaling with Large-scale Weak Supervision
24. ICML 2026Removing Sandbagging in LLMs by Training with Weak Supervision
25. ICML 2026The Label Horizon Paradox: Rethinking Supervision Targets in Financial Forecasting
26. ICML 2026TraCeS: Learning Per-Timestep Constraint-Violation Credit from Sparse Trajectory-Level Labels
27. ICML 2026Training Data Efficiency in Multimodal Process Reward Models
28. ICML 2026Trust Functions: Near Lossless Weak-to-Strong Generalization by Learning to Trust the Weak Teacher
29. ICML 2026Weak Diffusion Priors Can Still Achieve Strong Inverse-Problem Performance
30. ICML 2026Weak-to-Strong Generalization via Bregman Bias–Variance Decomposition
31. ICML 2026Weakly Supervised Cross-Modal Learning for 4D Radar Scene Flow Estimation
32. ICML 2026When Labelers Stay Silent: The Power of Ties in Cost-Effective Preference Learning
33. ICML 2026WS-GRPO: Weakly-Supervised Group-Relative Policy Optimization for Rollout-Efficient Reasoning
34. NeurIPS 2026Affine-invariant Cubic Newton with Weak Learners
35. NeurIPS 2026CAPER: Clause-Aligned Process Supervision for Text-to-SQL
36. NeurIPS 2026Clustering with Weak Distance Oracles
37. NeurIPS 2026From Patches to Trajectories: Privileged Process Supervision for Software-Engineering Agents
38. NeurIPS 2026Irreducible Supervision Enables Compositional Generalization in Post-Training
39. NeurIPS 2026Label-Free Consistency Correction for Weak-to-Strong Generalization
40. NeurIPS 2026Measuring Weak-to-Strong Legibility of Reasoning Models
41. NeurIPS 2026Neurosymbolic Object-Centric Learning with Distant Supervision
42. NeurIPS 2026Scalable Adaptation of 3D Geometric Foundation Models via Weak Supervision from Internet Video
43. NeurIPS 2026Uncertainty-Aware Probabilistic Constrained Clustering from Entangled Pairwise Supervision
44. NeurIPS 2026Uniform-in-Time Weak Propagation of Chaos in Shallow Neural Networks
45. NeurIPS 2026Your Teacher Can’t Help You Here: Combating Supervision Fidelity Decay in On-Policy Distillation
46. CVPR 2025Bridge the Gap: From Weak to Full Supervision for Temporal Action Localization with PseudoFormer
47. CVPR 2025Enhancing 3D Gaze Estimation in the Wild using Weak Supervision with Gaze Following Labels
48. CVPR 2025Instance-wise Supervision-level Optimization in Active Learning
49. CVPR 2025Masking meets Supervision: A Strong Learning Alliance
50. ICLR 2025Diff-Prompt: Diffusion-Driven Prompt Generator with Mask Supervision

常见问题

2023–2026 年四大会议一共有多少篇弱监督论文？

共 136 篇，按年份分别是 2023 年 34 篇、2024 年 27 篇、2025 年 30 篇、2026 年 45 篇。

NeurIPS 2026 有多少篇弱监督论文？

NeurIPS 2026 有 12 篇弱监督论文，上一届是 15 篇。

ICLR 2026 有多少篇弱监督论文？

ICLR 2026 有 8 篇弱监督论文，上一届是 9 篇。

ICML 2026 有多少篇弱监督论文？

ICML 2026 有 15 篇弱监督论文，上一届是 2 篇。

CVPR 2026 有多少篇弱监督论文？

CVPR 2026 有 10 篇弱监督论文，上一届是 4 篇。

哪个会议的弱监督论文最多？

NeurIPS，2023–2026 年共 42 篇。