长上下文论文趋势 2023–2026

主要结论

• 长上下文论文从 2023 年的 66 篇增长到 2026 年的 341 篇，三年间扩大到约 5.2 倍。
• NeurIPS 是发长上下文论文最多的会议，2023–2026 年共 298 篇，占四个会议总量的 36%。
• 交叉方向里涨得最快的是AI 智能体：同时属于两个主题的论文从 2023 年的 0 篇增至 2026 年的 45 篇。
• 在「长上下文与记忆」的 5 个主题里，它的 2026 年论文数排第 1；同组增长最快的检索记忆是 6.6 倍，长上下文是 5.2 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 2026Accelerating Long-Context LLM Prefill via Layer-wise Progressive Token Pruning in Local Deployment
• ICML 2026AdaSplash-2: Faster Differentiable Sparse Attention
• ICLR 2026An evolutionary perspective on modes of learning in Transformers
• CVPR 2026AR2-4FV Anchored Referring and Re-identification for Long-Term Grounding in Fixed-View
• NeurIPS 2025A Controllable Examination for Long-Context Language Models
• ICML 2025CateKV: On Sequential Consistency for Long-Context LLM Inference Acceleration
• ICLR 2025$\text{D}_{2}\text{O}$: Dynamic Discriminative Operations for Efficient Long-Context Inference of Large Language Models
• CVPR 2025BIMBA: Selective-Scan Compression for Long-Range Video Question Answering
• NeurIPS 2024[Re] CUDA: Curriculum of Data Augmentation for Long‐tailed Recognition
• ICML 2024Benchmarking and Building Long-Context Retrieval Models with LoCo and M2-BERT
• ICLR 2024CLEX: Continuous Length Extrapolation for Large Language Models
• CVPR 2024GigaTraj: Predicting Long-term Trajectories of Hundreds of Pedestrians in Gigapixel Complex Scenes
• NeurIPS 2023A Single 2D Pose with Context is Worth Hundreds for 3D Human Pose Estimation
• ICML 2023CAB: Comprehensive Attention Benchmarking on Long Sequence Modeling
• ICLR 2023$k$NN Prompting: Beyond-Context Learning with Calibration-Free Nearest Neighbor Inference
• CVPR 2023AutoAD: Movie Description in Context
• NeurIPS 2026AdaCal: Adaptive Calibration for Robust Sparse Attention in Long-Context LLMs
• ICML 2026Artificial Hippocampus Networks for Efficient Long-Context Modeling
• ICLR 2026AutoSP: Unlocking Long-Context LLM Training Via Compiler-Based Sequence Parallelism
• CVPR 2026AstraNav-Memory Contexts Compression for Long Memory

全部长上下文论文

共 341 篇，第 3 / 7 页

101. ICLR 2026SoLoPO: Unlocking Long-Context Capabilities in LLMs via Short-to-Long Preference Optimization
102. ICLR 2026Stacked from One: Multi-Scale Self-Injection for Context Window Extension
103. ICLR 2026Story-Iter: A Training-free Iterative Paradigm for Long Story Visualization
104. ICLR 2026SWINGARENA: Adversarial Programming Arena for Long-context GitHub Issue Solving
105. ICLR 2026Tactic: Adaptive Sparse Attention with Clustering and Distribution Fitting for Long-Context LLMs
106. ICLR 2026TAPTRv3: Spatial and Temporal Context Foster Robust Tracking of Any Point in Long Video
107. ICLR 2026Test-Time Training Done Right
108. ICLR 2026Transformers Learn Latent Mixture Models In-Context via Mirror Descent
109. ICLR 2026UltraLLaDA: Scaling the Context Length to 128K for Diffusion Large Language Models
110. ICLR 2026UltraMemV2: Memory Networks Scaling to 120B Parameters with Superior Long-Context Learning
111. ICLR 2026Understanding and Improving Length Generalization in Hierarchical Sparse Attention Models
112. ICLR 2026VideoChat-Flash: Hierarchical Compression for Long-Context Video Modeling
113. ICLR 2026When Does Divide and Conquer Work for Long Context LLM? A Noise Decomposition Framework
114. ICML 2026ACON: Optimizing Context Compression for Long-horizon LLM Agents
115. ICML 2026AdaSplash-2: Faster Differentiable Sparse Attention
116. ICML 2026Artificial Hippocampus Networks for Efficient Long-Context Modeling
117. ICML 2026ATLAS: Learning to Optimally Memorize the Context at Test Time
118. ICML 2026Auto-regressive In-context Demonstration Selection
119. ICML 2026BubbleSpec: Turning Long-Tail Bubbles into Speculative Rollout Drafts for Synchronous Reinforcement Learning
120. ICML 2026CLAA: Cross-Layer Attention Aggregation for Accelerating LLM Prefill
121. ICML 2026CoMem: Context Management with A Decoupled Long-Context Model
122. ICML 2026Context Forcing: Consistent Autoregressive Video Generation with Long Context
123. ICML 2026Context-Driven Incremental Compression for Multi-Turn Dialogue Generation
124. ICML 2026Context-free Recognition with Transformers
125. ICML 2026Context-level Language Modeling by Learning Predictive Context Embeddings
126. ICML 2026Doc-to-LoRA: Learning to Instantly Internalize Contexts
127. ICML 2026Dustin: Draft-Augmented Sparse Verification for Efficient Long-Context Generation with Speculative Decoding
128. ICML 2026EMFormer: Efficient Multi-Scale Transformer for Accumulative Context Weather Forecasting
129. ICML 2026EpiCache: Episodic KV Cache Management for Long-Term Conversation on Resource-Constrained Environments
130. ICML 2026Evaluating LLM Uncertainty in Long-Form Generation Using Deterministic Ground Truth
131. ICML 2026Fast KV Compaction via Attention Matching
132. ICML 2026Faster Than Flash: Exploiting Attention Sparsity for Efficient Long-Context Decoding
133. ICML 2026FlashBlock: Attention Caching for Efficient Long-Context Block Diffusion
134. ICML 2026ForesightKV: Optimizing KV Cache Eviction for Reasoning Models by Learning Long-Term Contribution
135. ICML 2026FRACTAL: State Space Model with Fractional Recurrent Architecture for Computational Temporal Analysis of Long Sequences
136. ICML 2026FunCQNet: A Functional Censored Quantile Neural Network for Predicting Long-Term Post-Transplant Kidney Survival
137. ICML 2026GradMem: Learning to Write Context into Memory with Test-Time Gradient Descent
138. ICML 2026Guidance: Sentence-Level Citation Enforcement via Prefix-Tail Guidance during LLM Decoding
139. ICML 2026HGMem: Hypergraph-based Working Memory to Improve Multi-step RAG for Long-Context Complex Relational Modeling
140. ICML 2026HiCI: Hierarchical Construction–Integration for Long-Context Attention
141. ICML 2026How Does the Pretraining Distribution Shape In-Context Learning? A Fundamental Trade-Off
142. ICML 2026How Many Different Outputs Can a Transformer Generate?
143. ICML 2026How Transformers Represent Hierarchies: A Local-to-Global Mechanism
144. ICML 2026IndexMem: Learned KV-Cache Eviction with Latent Memory for Long-Context LLM Inference
145. ICML 2026Induction Heads Interpolate N-Grams
146. ICML 2026InfoFlow KV: Information-Flow-Aware KV Recomputation for Long Context
147. ICML 2026KnapSpec: Self-Speculative Decoding via Adaptive Layer Selection as a Knapsack Problem
148. ICML 2026L-CUBE: Isolating Long-Context Capacity from Knowledge with Controllable Mutual Information Scaling
149. ICML 2026L-Drive: Beyond a Single Mapping—Latent Context Drives Time Series Forecasting
150. ICML 2026Learning When to Attend: Conditional Memory Access for Long-Context LLMs

常见问题

2023–2026 年四大会议一共有多少篇长上下文论文？

共 838 篇，按年份分别是 2023 年 66 篇、2024 年 176 篇、2025 年 255 篇、2026 年 341 篇。

NeurIPS 2026 有多少篇长上下文论文？

NeurIPS 2026 有 130 篇长上下文论文，上一届是 98 篇。

ICLR 2026 有多少篇长上下文论文？

ICLR 2026 有 82 篇长上下文论文，上一届是 87 篇。

ICML 2026 有多少篇长上下文论文？

ICML 2026 有 98 篇长上下文论文，上一届是 53 篇。

CVPR 2026 有多少篇长上下文论文？

CVPR 2026 有 31 篇长上下文论文，上一届是 17 篇。

哪个会议的长上下文论文最多？

NeurIPS，2023–2026 年共 298 篇。