多模态大模型论文趋势 2023–2026

主要结论

• 多模态大模型论文从 2023 年的 1,226 篇增长到 2026 年的 5,192 篇，三年间扩大到约 4.2 倍。
• NeurIPS 是发多模态大模型论文最多的会议，2023–2026 年共 3,319 篇，占四个会议总量的 30%。
• 交叉方向里涨得最快的是应用 · computer vision：同时属于两个主题的论文从 2023 年的 35 篇增至 2026 年的 596 篇。
• 在「前沿系统」的 4 个主题里，它的 2026 年论文数排第 1；同组增长最快的AI 智能体是 12 倍，多模态大模型是 4.2 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 2026$\text{A}^3$-VLA: Automatic Perception-Guided Attention Alignment for Robot Manipulation
• ICML 2026[CLS] is Not Enough: Multi-Label Recognition via Patch-Level Inference and Adaptive Aggregation
• ICLR 2026A Balanced Neuro-Symbolic Approach for Commonsense Abductive Logic
• CVPR 20263D Space as a Scratchpad for Editable Text-to-Image Generation
• NeurIPS 2025$\Delta \mathrm{Energy}$: Optimizing Energy Change During Vision-Language Alignment Improves both OOD Detection and OOD Generalization
• ICML 2025Adapting While Learning: Grounding LLMs for Scientific Problems with Tool Usage Adaptation
• ICLR 2025$\gamma-$MoD: Exploring Mixture-of-Depth Adaptation for Multimodal Large Language Models
• CVPR 20252DMamba: Efficient State Space Model for Image Representation with Applications on Giga-Pixel Whole Slide Image Classification
• NeurIPS 2024A Concept-Based Explainability Framework for Large Multimodal Models
• ICML 2024A Multimodal Automated Interpretability Agent
• ICLR 2024Alt-Text with Context: Improving Accessibility for Images on Twitter
• CVPR 2024Data-Efficient Multimodal Fusion on a Single GPU
• NeurIPS 20234M: Massively Multimodal Masked Modeling
• ICML 2023$\pi$-Tuning: Transferring Multimodal Foundation Models with Optimal Multi-task Interpolation
• ICLR 2023A Primal-Dual Framework for Transformers and Neural Networks
• CVPR 20233D Concept Learning and Reasoning From Multi-View Images
• NeurIPS 2026$\texttt{bispectrum}$: Selective $G$-Bispectra Made Practical
• ICML 2026$G^2$-Reader: Dual Evolving Graphs for Multimodal Document QA
• ICLR 2026A Comprehensive Information-Decomposition Analysis of Large Vision-Language Models
• CVPR 20263D-IDE 3D Implicit Depth Emergent

全部多模态大模型论文

共 1,401 篇，第 21 / 29 页

1001. ICLR 2026RL for Reasoning by Adaptively Revealing Rationales
1002. ICLR 2026RL makes MLLMs see better than SFT
1003. ICLR 2026RL Squeezes, SFT Expands: A Comparative Study of Reasoning LLMs
1004. ICLR 2026RLAD: Training LLMs to Discover Abstractions for Solving Reasoning Problems
1005. ICLR 2026RLAP-CLIP: Continual Multimodal Learning with Prototype Adaptation and Difficulty-Aware Routing
1006. ICLR 2026RMFlow: Refined Mean Flow by a Noise-Injection Step for Multimodal Generation
1007. ICLR 2026RoboCasa365: A Large-Scale Simulation Framework for Training and Benchmarking Generalist Robots
1008. ICLR 2026RoboInter: A Holistic Intermediate Representation Suite Towards Robotic Manipulation
1009. ICLR 2026RobotArena $\infty$: Unlimited Robot Benchmarking via Real-to-Sim Translation
1010. ICLR 2026Robotic Manipulation by Imitating Generated Videos Without Physical Demonstrations
1011. ICLR 2026Robust Fine-tuning of Vision-Language-Action Robot Policies via Parameter Merging
1012. ICLR 2026Robust Reward Modeling via Causal Rubrics
1013. ICLR 2026Robust Test-time Video-Text Retrieval: Benchmarking and Adapting for Query Shifts
1014. ICLR 2026RobustSpring: Benchmarking Robustness to Image Corruptions for Optical Flow, Scene Flow and Stereo
1015. ICLR 2026RoRE: Rotary Ray Embedding for Generalised Multi-Modal Scene Understanding
1016. ICLR 2026Rote Learning Considered Useful: Generalizing over Memorized Data in LLMs
1017. ICLR 2026Routing Manifold Alignment Improves Generalization of Mixture-of-Experts LLMs
1018. ICLR 2026Routing, Cascades, and User Choice for LLMs
1019. ICLR 2026ROVER: Benchmarking Reciprocal Cross-Modal Reasoning for Omnimodal Generation
1020. ICLR 2026RPM: Reasoning-Level Personalization for Black-Box Large Language Models
1021. ICLR 2026SABRE-FL: Selective and Accurate Backdoor Rejection for Federated Prompt Learning
1022. ICLR 2026SAES-SVD: Self-Adaptive Suppression of Accumulated and Local Errors for SVD-based LLM Compression
1023. ICLR 2026Safeguarding Multimodal Knowledge Copyright in the RAG-as-a-Service Environment
1024. ICLR 2026SafeMoE: Safe Fine-Tuning for MoE LLMs by Aligning Harmful Input Routing
1025. ICLR 2026Safety Mirage: How Spurious Correlations Undermine VLM Safety Fine-Tuning and Can Be Mitigated by Machine Unlearning
1026. ICLR 2026SAGE: Spatial-visual Adaptive Graph Exploration for Efficient Visual Place Recognition
1027. ICLR 2026SAM 3: Segment Anything with Concepts
1028. ICLR 2026Same Content, Different Representations: A Controlled Study for Table QA
1029. ICLR 2026Sample Lottery: Unsupervised Discovery of Critical Instances for LLM Reasoning
1030. ICLR 2026Sample Smart, Not Hard: Correctness-First Decoding for Better Reasoning in LLMs
1031. ICLR 2026Sapiens2
1032. ICLR 2026SASFT: Sparse Autoencoder-guided Supervised Finetuning to Mitigate Unexpected Code-Switching in LLMs
1033. ICLR 2026SC-Arena: A Natural Language Benchmark for Single-Cell Reasoning with Knowledge-Augmented Evaluation
1034. ICLR 2026Scalable Multilingual Multimodal Machine Translation with Speech-Text Fusion
1035. ICLR 2026ScaleCap: Scalable Image Captioning via Dual-Modality Debiasing
1036. ICLR 2026ScaleCUA: Scaling Open-Source Computer Use Agents with Cross-Platform Data
1037. ICLR 2026ScaleLong: A Multi-Timescale Benchmark for Long Video Understanding
1038. ICLR 2026Scaling Knowledge Editing in LLMs to 100,000 Facts with Neural KV Database
1039. ICLR 2026Scaling Laws Meet Model Architecture: Toward Inference-Efficient LLMs
1040. ICLR 2026Scaling up Memory for Robotic Control via Experience Retrieval
1041. ICLR 2026SceneCOT: Eliciting Chain-of-Thought Reasoning in 3D Scenes
1042. ICLR 2026SceneTransporter: Optimal Transport-Guided Compositional Latent Diffusion for Single-Image Structured 3D Scene Generation
1043. ICLR 2026SCI-Verifier: Scientific Verifier with Thinking
1044. ICLR 2026ScienceBoard: Evaluating Multimodal Autonomous Agents in Realistic Scientific Workflows
1045. ICLR 2026SciTS: Scientific Time Series Understanding and Generation with LLMs
1046. ICLR 2026SCUBA: Salesforce Computer Use Benchmark
1047. ICLR 2026Sculpting Subspaces: Constrained Full Fine-Tuning in LLMs for Continual Learning
1048. ICLR 2026Search Arena: Analyzing Search-Augmented LLMs
1049. ICLR 2026SeeDNorm: Self-Rescaled Dynamic Normalization
1050. ICLR 2026Seeing Across Views: Benchmarking Spatial Reasoning of Vision-Language Models in Robotic Scenes

常见问题

2023–2026 年四大会议一共有多少篇多模态大模型论文？

共 11,209 篇，按年份分别是 2023 年 1,226 篇、2024 年 1,716 篇、2025 年 3,075 篇、2026 年 5,192 篇。

NeurIPS 2026 有多少篇多模态大模型论文？

NeurIPS 2026 有 1,366 篇多模态大模型论文，上一届是 1,074 篇。

ICLR 2026 有多少篇多模态大模型论文？

ICLR 2026 有 1,401 篇多模态大模型论文，上一届是 816 篇。

ICML 2026 有多少篇多模态大模型论文？

ICML 2026 有 1,051 篇多模态大模型论文，上一届是 390 篇。

CVPR 2026 有多少篇多模态大模型论文？

CVPR 2026 有 1,374 篇多模态大模型论文，上一届是 795 篇。

哪个会议的多模态大模型论文最多？

NeurIPS，2023–2026 年共 3,319 篇。