扩散模型论文趋势 2023–2026
扩散、flow matching 与 score-based 生成
统计范围：NeurIPS、ICLR、ICML、CVPR 四个会议，2023–2026 年，共 6,900 篇被归入该主题的论文。
主要结论
- 扩散模型论文从 2023 年的 660 篇增长到 2026 年的 2,691 篇，三年间扩大到约 4.1 倍。
- NeurIPS 是发扩散模型论文最多的会议，2023–2026 年共 2,202 篇，占四个会议总量的 32%。
- 交叉方向里涨得最快的是应用 · chemistry physics and earth sciences：同时属于两个主题的论文从 2023 年的 14 篇增至 2026 年的 163 篇。
- 在「深度学习」的 11 个主题里，它的 2026 年论文数排第 1；同组增长最快的深度学习 · foundation models是 2.5 倍，扩散模型是 4.1 倍。
各会议论文数
按年份和会议统计
| 年份 | NeurIPS | ICLR | ICML | CVPR | 合计 |
|---|---|---|---|---|---|
| 2023 | 327 | 86 | 109 | 138 | 660 |
| 2024 | 520 | 260 | 195 | 496 | 1,471 |
| 2025 | 694 | 487 | 346 | 551 | 2,078 |
| 2026 | 661 | 694 | 728 | 608 | 2,691 |
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 2026$\mathbf{D^{3}S^{2}}$: Diffusion-Guided Dataset Distillation for Semantic Segmentation
- ICML 2026“Do Diffusion Models Dream of Electric Planes?” Discrete and Continuous Simulation-Based Inference for Aircraft Design
- ICLR 2026$\alpha$-DPO: Robust Preference Alignment for Diffusion Models via $\alpha$ Divergence
- CVPR 20262ndMatch Finetuning Pruned Diffusion Models via Second-Order Jacobian Matching
- NeurIPS 2025$\text{S}^2$Q-VDiT: Accurate Quantized Video Diffusion Transformer with Salient Data and Sparse Token Distillation
- ICML 2025A First-order Generative Bilevel Optimization Framework for Diffusion Models
- ICLR 2025$\text{I}^2\text{AM}$: Interpreting Image-to-Image Latent Diffusion Models via Bi-Attribution Maps
- CVPR 20253DEnhancer: Consistent Multi-View Diffusion for 3D Enhancement
- NeurIPS 2024$\text{Di}^2\text{Pose}$: Discrete Diffusion Model for Occluded 3D Human Pose Estimation
- ICML 2024A Dense Reward View on Aligning Text-to-Image Diffusion with Preference
- ICLR 2024$\infty$-Diff: Infinite Resolution Diffusion with Subsampled Mollified States
- CVPR 2024D^4: Dataset Distillation via Disentangled Diffusion Model
- NeurIPS 20233D molecule generation by denoising voxel grids
- ICML 2023A Critical Revisit of Adversarial Robustness in 3D Point Cloud Recognition with Diffusion-Driven Purification
- ICLR 2023(Certified!!) Adversarial Robustness for Free!
- CVPR 20233D Neural Field Generation Using Triplane Diffusion
- NeurIPS 2026$\text{PartConcepts}$: A Unified Mechanism for Fine-Grained Part Localization and Generation
- ICML 2026$\sigma$: Sigmoid Modulation for Ultra High Resolution Diffusion
- ICLR 2026$\textit{MADFormer}$: Mixed Autoregressive and Diffusion Transformers for Continuous Image Generation
- CVPR 20263D-LATTE Latent Space 3D Editing from Textual Instructions
全部扩散模型论文
共 2,202 篇，第 10 / 45 页
- NeurIPS 2026Projection Learning: A Principled Way to Overcome Memorization in Distribution Learning
- NeurIPS 2026Prompting Diffusion Models for Zero-Shot Instance Segmentation
- NeurIPS 2026Propagation of Chaos in Contextual Flow Maps
- NeurIPS 2026ProxyPose: 6-DoF Pose Tracking via Video-to-Video Translation
- NeurIPS 2026ProxySearch: Decoupled Inference-Time Scaling for Diffusion Models via Asymmetric Noise-Rank Transfer
- NeurIPS 2026PSD: Pushing the Pareto Frontier of Diffusion LLMs via Parallel Speculative Decoding
- NeurIPS 2026Q-ARVD: Quantizing Autoregressive Video Diffusion Models
- NeurIPS 2026Quantile-Coupled Flow Matching for Distributional Reinforcement Learning
- NeurIPS 2026Quantitative Assessment of Crystal Structure Prediction
- NeurIPS 2026Query Lower Bounds for Diffusion Sampling
- NeurIPS 2026RA-CFGCache: From Branch-Level Criteria to Guided-Risk Control under Classifier-Free Guidance
- NeurIPS 2026RadarFlowPose: Vision-Inspired Coarse-to-Fine Skeleton Refinement with Flow Matching
- NeurIPS 2026RB-LDC: Redundancy-Balanced Latent Coding for Robust Diffusion
- NeurIPS 2026Re-evaluating Confidence Remasking in Masked Diffusion Language Models
- NeurIPS 2026Read It Back: Pretrained MLLMs Are Zero-shot Reward Models for Text-to-Image Generation
- NeurIPS 2026Reading Positional Coupling in Transformers with Diffusion Scores
- NeurIPS 2026Reasoning with Undecoded Tokens in Diffusion Language Models
- NeurIPS 2026Reconsidering Positional Supervision in Masked Diffusion Language Model Training
- NeurIPS 2026Reconstruction or Semantics? What Makes a Latent Space Useful for Robotic World Models
- NeurIPS 2026Recursively Trained Diffusion Models: Limiting Collapse Distribution and Spectral Characterization
- NeurIPS 2026RefDecoder: Enhancing Visual Generation with Conditional Video Decoding
- NeurIPS 2026Refining Compositional Diffusion for Reliable Long-Horizon Planning
- NeurIPS 2026Reflected Schrödinger Bridge Matching
- NeurIPS 2026ReFlex: Faithful Panorama Reconstruction from a Single Image with Reflections
- NeurIPS 2026ReFPO: Reflow Regularization for Flow Matching Policy Gradients
- NeurIPS 2026ReGuidance: Diffusion Steering with Strong Latent Initializations Solves Hard Inverse Problems
- NeurIPS 2026Reinforce Adjoint Matching: Scaling RL Post-Training of Diffusion and Flow Models
- NeurIPS 2026Reinforcement learning enhanced flow matching for reference-based tumor generation on unpaired CT images
- NeurIPS 2026Reinforcement Learning for Diffusion LLMs with Entropy-Guided Step Selection and Stepwise Advantages
- NeurIPS 2026Reject, Resample, Repeat: Understanding Parallel Reasoning in Language Model Inference
- NeurIPS 2026Reliability-Coupled Manifold-Aware Diffusion for Missing-Modality Inference
- NeurIPS 2026Renoise Consistency: Unlocking Efficient Self-Correction for Diffusion Large Language Models
- NeurIPS 2026RePiD: Efficient Recursive Pixel-Space Diffusion via Hierarchical Patch Denoising
- NeurIPS 2026RepoScope: Bridging Physical Structure and Logical Flow for Intent-Driven Repository Documentation
- NeurIPS 2026Representation Preconditioning for Efficient Diffusion Model Training
- NeurIPS 2026Repurposing Video Diffusion Transformers for Cross-View Temporal Object Correspondence
- NeurIPS 2026ResFusion: Medical Image Fusion Driven by Implicit-Forward Diffusion and Time-aware Joint Optimization
- NeurIPS 2026Response Time Enhances Alignment with Heterogeneous Preferences
- NeurIPS 2026Rethinking Causal Action Tokenization with Conditional Annealing in Flow Matching
- NeurIPS 2026Rethinking Diffusion Decoding via Structural Commitment
- NeurIPS 2026Rethinking Knowledge Distillation for Diffusion Language Models
- NeurIPS 2026Rethinking Learning from Label Proportions via Moment Matching
- NeurIPS 2026Rethinking Memorization–Generalization Trade-Off in Generative Models
- NeurIPS 2026Rethinking Vector Field Learning for Generative Segmentation
- NeurIPS 2026Reusable Conditional Resampling via Flow Matching for Constraint-Based Causal Discovery
- NeurIPS 2026Reversing the Roles of Signal and Noise: Modeling Noise in Astronomical Images without Paired Data
- NeurIPS 2026Revisiting The Power of Closed-Form: Robust Deep Image Prototype Discovery via Scale Mixtures
- NeurIPS 2026Reweighted Flow Matching via Unbalanced Optimal Transport for Label-free Long-tailed Generation
- NeurIPS 2026RIFLE: Removal of Image Flicker-Banding via Latent Diffusion Enhancement
- NeurIPS 2026RoboExo: Structure-Guided Wrist-to-Exocentric Video Generation for Scalable Robot Learning
在 Claude、Cursor 里直接问这个趋势
接上 Paperlist MCP 后，写综述、挑方向或准备投稿时，不用离开对话框就能拿到这页的数字：每年各会议的论文数、增长倍数、涨得最快的子方向和代表论文，AI 会按你的问题整理。
可以直接复制这句话问 AI
用 paperlist 查扩散模型在 NeurIPS 2023–2026 的论文趋势，哪一年涨得最快？
AI 会调用的工具：get_topic_stats {"topic": "diffusion-models"}
常见问题
- 2023–2026 年四大会议一共有多少篇扩散模型论文？
- 共 6,900 篇，按年份分别是 2023 年 660 篇、2024 年 1,471 篇、2025 年 2,078 篇、2026 年 2,691 篇。
- NeurIPS 2026 有多少篇扩散模型论文？
- NeurIPS 2026 有 661 篇扩散模型论文，上一届是 694 篇。
- ICLR 2026 有多少篇扩散模型论文？
- ICLR 2026 有 694 篇扩散模型论文，上一届是 487 篇。
- ICML 2026 有多少篇扩散模型论文？
- ICML 2026 有 728 篇扩散模型论文，上一届是 346 篇。
- CVPR 2026 有多少篇扩散模型论文？
- CVPR 2026 有 608 篇扩散模型论文，上一届是 551 篇。
- 哪个会议的扩散模型论文最多？
- NeurIPS，2023–2026 年共 2,202 篇。
「深度学习」下的其他主题
- 深度学习 · generative models and autoencoders 4,314
- 深度学习 · large language models 2,334
- 深度学习 · self supervised learning 1,958
- 深度学习 · graph neural networks 1,648
- 深度学习 · optimization 953
- 深度学习 · representation learning 802
- 深度学习 · probabilistic methods 637
- 深度学习 · foundation models 536
- 深度学习 · causal inference 316
- 深度学习 · transfer learning 197