合成数据论文趋势 2023–2026

主要结论

• 合成数据论文从 2023 年的 54 篇降到 2026 年的 50 篇，热度在回落。
• NeurIPS 是发合成数据论文最多的会议，2023–2026 年共 64 篇，占四个会议总量的 38%。
• 交叉方向里涨得最快的是强化学习：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「联邦与隐私」的 5 个主题里，它的 2026 年论文数排第 4；同组增长最快的安全聚合是 1.8 倍，合成数据是 0.9 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 2026Learning with Synthetic Data via SGD in High-Dimensional Linear Regression
• ICML 2026A Statistical Framework for Analyzing Specification Resistance to Learnware-Inversion Risks
• ICLR 2026Synthetic Bootstrapped Pretraining
• CVPR 2026UnReflectAnything RGB-Only Highlight Removal by Rendering Synthetic Specular Supervision
• NeurIPS 2025Achilles' Heel of Mamba: Essential difficulties of the Mamba architecture demonstrated by synthetic data
• ICML 2025Synthesizing Privacy-Preserving Text Data via Finetuning *without* Finetuning Billion-Scale LLMs
• ICLR 2025Fictitious Synthetic Data Can Improve LLM Factuality via Prerequisite Learning
• CVPR 2025Synthetic Visual Genome
• NeurIPS 2024DECO-Bench: Unified Benchmark for Decoupled Task-Agnostic Synthetic Data Release
• ICML 2024Compositional Capabilities of Autoregressive Transformers: A Study on Synthetic, Interpretable Tasks
• ICLR 2024Differentially Private Synthetic Data via Foundation Model APIs 1: Images
• CVPR 2024Discontinuity-preserving Normal Integration with Auxiliary Edges
• NeurIPS 2023Can You Rely on Your Model Evaluation? Improving Model Evaluation with Synthetic Test Data
• ICML 2023Data Structures for Density Estimation
• ICLR 2023Statistical Theory of Differentially Private Marginal-based Data Synthesis Algorithms
• CVPR 20233D-Aware Facial Landmark Detection via Multi-View Consistent Training on Synthetic Data
• NeurIPS 2026StereoPep: Do Molecular Models Understand Stereochemistry? A Benchmark on Synthetic Diastereomeric Peptides
• ICML 2026FedUSD: Unbiased Synthetic Data for Federated Learning
• ICLR 2026High-dimensional Analysis of Synthetic Data Selection
• CVPR 2026AHS Adaptive Head Synthesis via Synthetic Data Augmentations

全部合成数据论文

共 169 篇，第 1 / 4 页

1. CVPR 2026AHS Adaptive Head Synthesis via Synthetic Data Augmentations
2. CVPR 2026BarbieGait An Identity-Consistent Synthetic Human Dataset with Versatile Cloth-Changing for
3. CVPR 2026CaricHarmony Contrastive Diffusion Paths for Identity-Preserving Caricature Synthesis
4. CVPR 2026Exemplar-Free Class Incremental Learning via Preserving Class-Discriminative Structure
5. CVPR 2026Layer Consistency Matters Elegant Latent Transition Discrepancy for Generalizable Synthetic
6. CVPR 2026Learning from Synthetic Data via Provenance-Based Input Gradient Guidance
7. CVPR 2026Learning to Generate Highly Dynamic Videos using Synthetic Motion Data
8. CVPR 2026PLACID Identity-Preserving Multi-Object Compositing via Video Diffusion with Synthetic Trajectories
9. CVPR 2026PrivateEyes Gaze-Preserving Anonymization for Data Sharing
10. CVPR 2026Reinforcement-Guided Synthetic Data Generation for Privacy-Sensitive Identity Recognition
11. CVPR 2026Role-SynthCLIP A Role-Play Driven Diverse Synthetic Data Approach
12. CVPR 2026UnReflectAnything RGB-Only Highlight Removal by Rendering Synthetic Specular Supervision
13. CVPR 2026Vanast Virtual Try-On with Human Image Animation via Synthetic Triplet
14. CVPR 2026Whats Wrong with Synthetic Data for Scene Text Recognition A
15. ICLR 2026Achieving Expert-Level Agent from Foundation Model via Complexity Curriculum Reinforcement Learning with Synthetic Data
16. ICLR 2026Entropy-preserving reinforcement learning
17. ICLR 2026High-dimensional Analysis of Synthetic Data Selection
18. ICLR 2026Synthetic Bootstrapped Pretraining
19. ICML 2026$\texttt{FHAIM}$: Fully Homomorphic AIM for Private Tabular Synthetic Data Generation
20. ICML 2026A Statistical Framework for Analyzing Specification Resistance to Learnware-Inversion Risks
21. ICML 2026An Information-Theoretic Criterion for Efficient Data Synthesis
22. ICML 2026Differentially Private Synthetic Data via APIs 4: Tabular Data
23. ICML 2026FedUSD: Unbiased Synthetic Data for Federated Learning
24. ICML 2026Geometry-Aware Tabular Diffusion
25. ICML 2026Identifying Learnwares via Reduced Neural Conditional Mean Embedding
26. ICML 2026Local Redundancy: An Information-Theoretic Measure of Plasticity from Synthetic Memorization
27. ICML 2026Optimal Domain-Aware Privacy Mechanisms for Synthetic Data Generation
28. ICML 2026PluRel: Synthetic Data unlocks Scaling Laws for Relational Foundation Models
29. ICML 2026R1-SyntheticVL: Is Synthetic Data from Generative Models Ready for Multimodal Large Language Model?
30. ICML 2026Relational In-Context Learning via Synthetic Pre-training with Structural Prior
31. ICML 2026TexEditor: Structure-Preserving Text-Driven Texture Editing
32. ICML 2026Unlearning with Asymmetric Sources: Improved Unlearning-Utility Trade-off with Public Data
33. NeurIPS 2026AnyHand: A Large-Scale Synthetic Dataset for RGB(-D) Hand Pose Estimation
34. NeurIPS 2026Breaking the Synthesis Barrier for AI-Designed DNA Libraries
35. NeurIPS 2026Evaluating Synthetic ECG Pretraining: When Can Patient-Free Simulators Substitute for Real ECG Data?
36. NeurIPS 2026Is Backpropagation Optimal? When Synthetic Gradients Improve Sample Efficiency
37. NeurIPS 2026It's All Training: A Fully Synthetic Single-Stage Recipe for LLMs
38. NeurIPS 2026Learning with Synthetic Data via SGD in High-Dimensional Linear Regression
39. NeurIPS 2026Load Balancing Mixture of Experts with Similarity Preserving Routers
40. NeurIPS 2026Measuring What Matters: Synthetic Benchmarks for Concept Bottleneck Models
41. NeurIPS 2026Optimal Real-Data Allocation for Synthetic-Data-Augmented Inference
42. NeurIPS 2026Persona Generators: Generating Diverse Synthetic Personas for Arbitrary Contexts
43. NeurIPS 2026Pion: A Spectrum-Preserving Optimizer via Orthogonal Equivalence Transformation
44. NeurIPS 2026Privacy Amplification Persists under Unlimited Synthetic Data Release
45. NeurIPS 2026Quest: Training Frontier Deep Research Agents with Fully Synthetic Tasks
46. NeurIPS 2026SLS-Bench: A Benchmark for Incident Log Summarization with Synthetic Observability Data
47. NeurIPS 2026StereoPep: Do Molecular Models Understand Stereochemistry? A Benchmark on Synthetic Diastereomeric Peptides
48. NeurIPS 2026Synthetic Anchor-Assisted Prototype Alignment for Heterogeneous Federated Learning
49. NeurIPS 2026Synthetic Pre-Pre-Training Improves Language Model Robustness to Noisy Pre-Training Data
50. NeurIPS 2026Targeted Synthetic Control Method

常见问题

2023–2026 年四大会议一共有多少篇合成数据论文？

共 169 篇，按年份分别是 2023 年 54 篇、2024 年 33 篇、2025 年 32 篇、2026 年 50 篇。

NeurIPS 2026 有多少篇合成数据论文？

NeurIPS 2026 有 18 篇合成数据论文，上一届是 8 篇。

ICLR 2026 有多少篇合成数据论文？

ICLR 2026 有 4 篇合成数据论文，上一届是 7 篇。

ICML 2026 有多少篇合成数据论文？

ICML 2026 有 14 篇合成数据论文，上一届是 9 篇。

CVPR 2026 有多少篇合成数据论文？

CVPR 2026 有 14 篇合成数据论文，上一届是 8 篇。

哪个会议的合成数据论文最多？

NeurIPS，2023–2026 年共 64 篇。