上下文压缩论文趋势 2023–2026

主要结论

• 上下文压缩论文从 2023 年的 39 篇增长到 2026 年的 136 篇，三年间扩大到约 3.5 倍。
• NeurIPS 是发上下文压缩论文最多的会议，2023–2026 年共 123 篇，占四个会议总量的 34%。
• 交叉方向里涨得最快的是深度学习 · large language models：同时属于两个主题的论文从 2023 年的 1 篇增至 2026 年的 13 篇。
• 在「长上下文与记忆」的 5 个主题里，它的 2026 年论文数排第 4；同组增长最快的检索记忆是 6.6 倍，上下文压缩是 3.5 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 2026A Free Lunch in LLM Compression: Revisiting Retraining after Pruning
• ICML 2026A Unified Sparse Attention via Multi-Granularity Compression
• ICLR 2026An evolutionary perspective on modes of learning in Transformers
• CVPR 2026AstraNav-Memory Contexts Compression for Long Memory
• NeurIPS 2025AdmTree: Compressing Lengthy Context with Adaptive Semantic Trees
• ICML 2025Does learning the right latent variables necessarily improve in-context learning?
• ICLR 2025Approaching Rate-Distortion Limits in Neural Compression with Lattice Transform Coding
• CVPR 2025ECVC: Exploiting Non-Local Correlations in Multiple Frames for Contextual Video Compression
• NeurIPS 2024StreamingDialogue: Prolonged Dialogue Learning via Long Context Compression with Minimal Losses
• ICML 2024Agnostic Sample Compression Schemes for Regression
• ICLR 2024Compressed Context Memory for Online Language Model Interaction
• CVPR 2024Diversity-aware Channel Pruning for StyleGAN Compression
• NeurIPS 2023Do Not Marginalize Mechanisms, Rather Consolidate!
• ICML 2023Private Federated Learning with Autotuned Compression
• ICLR 2023$k$NN Prompting: Beyond-Context Learning with Calibration-Free Nearest Neighbor Inference
• CVPR 2023AutoAD: Movie Description in Context
• NeurIPS 2026A Self-Evolving Framework for Efficient Terminal Agents via Observational Context Compression
• ICML 2026Context-Driven Incremental Compression for Multi-Turn Dialogue Generation
• ICLR 2026Autoencoding-Free Context Compression for LLMs via Contextual Semantic Anchors
• CVPR 2026Featurising Pixels from Dynamic 3D Scenes with Linear In-Context Learners

全部上下文压缩论文

共 109 篇，第 2 / 3 页

51. ICLR 2025You Only Prune Once: Designing Calibration-Free Model Compression With Policy Learning
52. ICML 2025CommVQ: Commutative Vector Quantization for KV Cache Compression
53. ICML 2025Compression via Pre-trained Transformers: A Study on Byte-Level Multimodal Data
54. ICML 2025DLP: Dynamic Layerwise Pruning in Large Language Models
55. ICML 2025Does learning the right latent variables necessarily improve in-context learning?
56. ICML 2025Efficient LiDAR Reflectance Compression via Scanning Serialization
57. ICML 2025Evolving Prompts In-Context: An Open-ended, Self-replicating Perspective
58. ICML 2025EvoPress: Accurate Dynamic Model Compression via Evolutionary Search
59. ICML 2025In-Context Deep Learning via Transformer Models
60. ICML 2025In-Context Learning and Occam's Razor
61. ICML 2025Iterative Vectors: In-Context Gradient Steering without Backpropagation
62. ICML 2025Learning In-context $n$-grams with Transformers: Sub-$n$-grams Are Near-Stationary Points
63. ICML 2025Lexico: Extreme KV Cache Compression via Sparse Coding over Universal Dictionaries
64. ICML 2025NExtLong: Toward Effective Long-Context Training without Long Documents
65. ICML 2025ParallelComp: Parallel Long-Context Compressor for Length Extrapolation
66. ICML 2025Prompt-based Depth Pruning of Large Language Models
67. ICML 2025Retraining-free Merging of Sparse MoE via Hierarchical Clustering
68. ICML 2025RocketKV: Accelerating Long-Context LLM Inference via Two-Stage KV Cache Compression
69. ICML 2025SkipGPT: Each Token is One of a Kind
70. ICML 2025Training Dynamics of In-Context Learning in Linear Attention
71. ICML 2025Weight matrices compression based on PDB model in deep neural networks
72. NeurIPS 2025AdmTree: Compressing Lengthy Context with Adaptive Semantic Trees
73. NeurIPS 2025AttentionPredictor: Temporal Patterns Matter for KV Cache Compression
74. NeurIPS 2025Binary Quadratic Quantization: Beyond First-Order Quantization for Real-Valued Matrix Compression
75. NeurIPS 2025Breaking the Compression Ceiling: Data-Free Pipeline for Ultra-Efficient Delta Compression
76. NeurIPS 2025ChunkKV: Semantic-Preserving KV Cache Compression for Efficient Long-Context LLM Inference
77. NeurIPS 2025COALA: Numerically Stable and Efficient Framework for Context-Aware Low-Rank Approximation
78. NeurIPS 2025Conditional Distribution Compression via the Kernel Conditional Mean Embedding
79. NeurIPS 2025Constant Bit-size Transformers Are Turing Complete
80. NeurIPS 2025DeepHalo: A Neural Choice Model with Controllable Context Effects
81. NeurIPS 2025Disentangling Latent Shifts of In-Context Learning with Weak Supervision
82. NeurIPS 2025Don't Just Chase “Highlighted Tokens” in MLLMs: Revisiting Visual Holistic Context Retention
83. NeurIPS 2025DUO: No Compromise to Accuracy Degradation
84. NeurIPS 2025Efficient Prompt Compression with Evaluator Heads for Long-Context Transformer Inference
85. NeurIPS 2025End-to-End Low-Light Enhancement for Object Detection with Learned Metadata from RAWs
86. NeurIPS 2025FP4 All the Way: Fully Quantized Training of Large Language Models
87. NeurIPS 2025From Softmax to Score: Transformers Can Effectively Implement In-Context Denoising Steps
88. NeurIPS 2025Homogeneous Keys, Heterogeneous Values: Exploiting Local KV Cache Asymmetry for Long-Context LLMs
89. NeurIPS 2025In-Context Compositional Learning vis Sparse Coding Transformer
90. NeurIPS 2025InfiniPot-V: Memory-Constrained KV Cache Compression for Streaming Video Understanding
91. NeurIPS 2025KeyDiff: Key Similarity-Based KV Cache Eviction for Long-Context LLM Inference in Resource-Constrained Environments
92. NeurIPS 2025KVzip: Query-Agnostic KV Cache Compression with Context Reconstruction
93. NeurIPS 2025Language Models (Mostly) Know When to Stop Reading
94. NeurIPS 2025Large Language Models for Lossless Image Compression: Next-Pixel Prediction in Language Space is All You Need
95. NeurIPS 2025LittleBit: Ultra Low-Bit Quantization via Latent Factorization
96. NeurIPS 2025MoRIC: A Modular Region-based Implicit Codec for Image Compression
97. NeurIPS 2025Neural Attention Search
98. NeurIPS 2025Neural B-frame Video Compression with Bi-directional Reference Harmonization
99. NeurIPS 2025Overcoming Long Context Limitations of State Space Models via Context Dependent Sparse Attention
100. NeurIPS 2025Scale-invariant attention

常见问题

2023–2026 年四大会议一共有多少篇上下文压缩论文？

共 365 篇，按年份分别是 2023 年 39 篇、2024 年 81 篇、2025 年 109 篇、2026 年 136 篇。

NeurIPS 2026 有多少篇上下文压缩论文？

NeurIPS 2026 有 50 篇上下文压缩论文，上一届是 38 篇。

ICLR 2026 有多少篇上下文压缩论文？

ICLR 2026 有 28 篇上下文压缩论文，上一届是 32 篇。

ICML 2026 有多少篇上下文压缩论文？

ICML 2026 有 39 篇上下文压缩论文，上一届是 20 篇。

CVPR 2026 有多少篇上下文压缩论文？

CVPR 2026 有 19 篇上下文压缩论文，上一届是 19 篇。

哪个会议的上下文压缩论文最多？

NeurIPS，2023–2026 年共 123 篇。