合成数据生成论文趋势 2023–2026

主要结论

• 合成数据生成论文从 2023 年的 72 篇增长到 2026 年的 96 篇，三年间扩大到约 1.3 倍。
• NeurIPS 是发合成数据生成论文最多的会议，2023–2026 年共 114 篇，占四个会议总量的 41%。
• 交叉方向里涨得最快的是扩散模型：同时属于两个主题的论文从 2023 年的 6 篇增至 2026 年的 20 篇。
• 在「以数据为中心 AI」的 5 个主题里，它的 2026 年论文数排第 1；同组增长最快的弱监督是 1.3 倍，合成数据生成是 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 2026Breaking the Quality–Privacy Tradeoff in Tabular Data Generation via In-Context Learning
• ICML 2026Active Tabular Augmentation via Policy-Guided Diffusion Inpainting
• ICLR 2026Critic–Adviser–Reviser Cyclic Refinement: Towards High-Quality EMR Corpus Generation with LLMs
• CVPR 2026Beyond Objects Contextual Synthetic Data Generation for Fine-Grained Classification
• NeurIPS 2025Achilles' Heel of Mamba: Essential difficulties of the Mamba architecture demonstrated by synthetic data
• ICML 2025Synthetic Text Generation for Training Large Language Models via Gradient Matching
• ICLR 2025Scaling Instruction-tuned LLMs to Million-token Contexts via Hierarchical Synthetic Data Generation
• CVPR 2025Synthetic Visual Genome
• NeurIPS 2024DECO-Bench: Unified Benchmark for Decoupled Task-Agnostic Synthetic Data Release
• ICML 2024CaRiNG: Learning Temporal Causal Representation under Non-Invertible Generation Process
• ICLR 2024A Linear Algebraic Framework for Counterfactual Generation
• CVPR 2024Learning from Synthetic Human Group Activities
• NeurIPS 2023[Re] $\mathcal{G}$-Mixup: Graph Data Augmentation for Graph Classification
• ICML 2023Data Feedback Loops: Model-driven Amplification of Dataset Biases
• ICLR 2023Synthetic Data Generation of Many-to-Many Datasets via Random Graph Generation
• CVPR 2023Fake It Till You Make It: Learning Transferable Representations From Synthetic ImageNet Clones
• NeurIPS 2026Learning with Synthetic Data via SGD in High-Dimensional Linear Regression
• ICML 2026Counterfactual Residual Data Augmentation for Regression
• ICLR 2026Making, Not Taking, the Best of N
• CVPR 2026Order Matters 3D Shape Generation from Sequential VR Sketches

全部合成数据生成论文

共 277 篇，第 3 / 6 页

101. CVPR 2025Enhancing Virtual Try-On with Synthetic Pairs and Error-Aware Noise Scheduling
102. CVPR 2025From Sparse Signal to Smooth Motion: Real-Time Motion Generation with Rolling Prediction Models
103. CVPR 2025MIDI: Multi-Instance Diffusion for Single Image to 3D Scene Generation
104. CVPR 2025Motion Modes: What Could Happen Next?
105. CVPR 2025MultiMorph: On-demand Atlas Construction
106. CVPR 2025S2D-LFE: Sparse-to-Dense Light Field Event Generation
107. CVPR 2025SeaLion: Semantic Part-Aware Latent Point Diffusion Models for 3D Generation
108. CVPR 2025Synthetic Visual Genome
109. CVPR 2025SynthLight: Portrait Relighting with Diffusion Model by Learning to Re-render Synthetic Faces
110. CVPR 2025Transfer Your Perspective: Controllable 3D Generation from Any Viewpoint in a Driving Scene
111. ICLR 2025API Pack: A Massive Multi-Programming Language Dataset for API Call Generation
112. ICLR 2025Beyond the convexity assumption: Realistic tabular data generation under quantifier-free real linear constraints
113. ICLR 2025Conditional Diffusion with Ordinal Regression: Longitudinal Data Generation for Neurodegenerative Disease Studies
114. ICLR 2025Continuous Diffusion for Mixed-Type Tabular Data
115. ICLR 2025CraftRTL: High-quality Synthetic Data Generation for Verilog Code Models with Correct-by-Construction Non-Textual Representations and Targeted Code Repair
116. ICLR 2025CTSyn: A Foundation Model for Cross Tabular Data Generation
117. ICLR 2025DataEnvGym: Data Generation Agents in Teacher Environments with Student Feedback
118. ICLR 2025DataGen: Unified Synthetic Dataset Generation via Large Language Models
119. ICLR 2025Fictitious Synthetic Data Can Improve LLM Factuality via Prerequisite Learning
120. ICLR 2025Gaussian Mixture Counterfactual Generator
121. ICLR 2025MindSimulator: Exploring Brain Concept Localization via Synthetic fMRI
122. ICLR 2025Personalized Representation from Personalized Generation
123. ICLR 2025Scaling Instruction-tuned LLMs to Million-token Contexts via Hierarchical Synthetic Data Generation
124. ICLR 2025Single Teacher, Multiple Perspectives: Teacher Knowledge Augmentation for Enhanced Knowledge Distillation
125. ICLR 2025Synthetic continued pretraining
126. ICLR 2025The Power of LLM-Generated Synthetic Data for Stance Detection in Online Political Discussions
127. ICLR 2025Towards a Theoretical Understanding of Synthetic Data in LLM Post-Training: A Reverse-Bottleneck Perspective
128. ICLR 2025Vec2Face: Scaling Face Dataset Generation with Loosely Constrained Vectors
129. ICML 2025BAnG: Bidirectional Anchored Generation for Conditional RNA Design
130. ICML 2025BOOD: Boundary-based Out-Of-Distribution Data Generation
131. ICML 2025Breaking the Barrier of Hard Samples: A Data-Centric Approach to Synthetic Data for Medical Tasks
132. ICML 2025Learning-Order Autoregressive Models with Application to Molecular Graph Generation
133. ICML 2025Offline-to-Online Reinforcement Learning with Classifier-Free Diffusion Generation
134. ICML 2025SK-VQA: Synthetic Knowledge Generation at Scale for Training Context-Augmented Multimodal LLMs
135. ICML 2025Synthetic Face Datasets Generation via Latent Space Exploration from Brownian Identity Diffusion
136. ICML 2025Synthetic Text Generation for Training Large Language Models via Gradient Matching
137. ICML 2025TabSDS: a Lightweight, Fully Non-Parametric, and Model Free Approach for Generating Synthetic Tabular Data
138. NeurIPS 2025Achilles' Heel of Mamba: Essential difficulties of the Mamba architecture demonstrated by synthetic data
139. NeurIPS 2025Amortized Active Generation of Pareto Sets
140. NeurIPS 2025AR-RAG: Autoregressive Retrieval Augmentation for Image Generation
141. NeurIPS 2025AugGen: Synthetic Augmentation using Diffusion Models Can Improve Recognition
142. NeurIPS 2025ChemOrch: Empowering LLMs with Chemical Intelligence via Groundbreaking Synthetic Instructions
143. NeurIPS 2025Constrained Posterior Sampling: Time Series Generation with Hard Constraints
144. NeurIPS 2025Diffusing DeBias: Synthetic Bias Amplification for Model Debiasing
145. NeurIPS 2025FerretNet: Efficient Synthetic Image Detection via Local Pixel Dependencies
146. NeurIPS 2025Forging Time Series with Language: A Large Language Model Approach to Synthetic Data Generation
147. NeurIPS 2025Improving the Generation and Evaluation of Synthetic Data for Downstream Medical Causal Inference
148. NeurIPS 2025Increasing the Utility of Synthetic Images through Chamfer Guidance
149. NeurIPS 2025KGGen: Extracting Knowledge Graphs from Plain Text with Language Models
150. NeurIPS 2025MesaTask: Towards Task-Driven Tabletop Scene Generation via 3D Spatial Reasoning

常见问题

2023–2026 年四大会议一共有多少篇合成数据生成论文？

共 277 篇，按年份分别是 2023 年 72 篇、2024 年 41 篇、2025 年 68 篇、2026 年 96 篇。

NeurIPS 2026 有多少篇合成数据生成论文？

NeurIPS 2026 有 24 篇合成数据生成论文，上一届是 27 篇。

ICLR 2026 有多少篇合成数据生成论文？

ICLR 2026 有 26 篇合成数据生成论文，上一届是 18 篇。

ICML 2026 有多少篇合成数据生成论文？

ICML 2026 有 21 篇合成数据生成论文，上一届是 9 篇。

CVPR 2026 有多少篇合成数据生成论文？

CVPR 2026 有 25 篇合成数据生成论文，上一届是 14 篇。

哪个会议的合成数据生成论文最多？

NeurIPS，2023–2026 年共 114 篇。