深度学习 · large language models论文趋势 2023–2026

主要结论

• 深度学习 · large language models论文从 2023 年的 488 篇增长到 2026 年的 1,131 篇，三年间扩大到约 2.3 倍。
• ICML 是发深度学习 · large language models论文最多的会议，2023–2026 年共 1,608 篇，占四个会议总量的 69%。
• 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 5 篇增至 2026 年的 184 篇。
• 在「深度学习」的 11 个主题里，它的 2026 年论文数排第 3；同组增长最快的扩散模型是 4.1 倍，深度学习 · large language models是 2.3 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

• ICML 2026“very likely” Means “uncertain”? How LLMs Diverge from Humans in Linguistic Uncertainty Quantification
• ICML 2025(How) Do Language Models Track State?
• ICLR 2025AdvWave: Stealthy Adversarial Jailbreak Attack against Large Audio-Language Models
• NeurIPS 2024$\beta$-DPO: Direct Preference Optimization with Dynamic $\beta$
• NeurIPS 20233D-LLM: Injecting the 3D World into Large Language Models
• ICML 2023Are Large Kernels Better Teachers than Transformers for ConvNets?
• ICML 2026$f$-Divergence Self-Play for Tabular Anomaly Detection via Large Language Models
• ICML 2025A Tale of Two Structures: Do LLMs Capture the Fractal Complexity of Language?
• ICLR 2025Emergence of a High-Dimensional Abstraction Phase in Language Transformers
• NeurIPS 2024$\textit{Trans-LoRA}$: towards data-free Transferable Parameter Efficient Finetuning
• NeurIPS 2023A fast heuristic to optimize time-space tradeoff for large models
• ICML 2023Arithmetic Sampling: Parallel Diverse Decoding for Large Language Models
• ICML 2026$L^3$: Large Lookup Layers
• ICML 2025Accelerating LLM Inference with Lossless Speculative Decoding Algorithms for Heterogeneous Vocabularies
• ICLR 2025Holistically Evaluating the Environmental Impact of Creating Language Models
• NeurIPS 2024A distributional simplicity bias in the learning dynamics of transformers
• NeurIPS 2023A Pseudo-Semantic Loss for Autoregressive Models with Logical Constraints
• ICML 2023Auto-Differentiation of Relational Computations for Very Large Scale Machine Learning
• ICML 2026$V_0$: A Generalist Value Model for Any Policy at State Zero
• ICML 2025Accelerating Unbiased LLM Evaluation via Synthetic Feedback

全部深度学习 · large language models论文

共 1,131 篇，第 23 / 23 页

1101. ICML 2026What LLMs Explain Is Not What They Believe: Evaluating Explanation Sufficiency Under Models' Own Input Beliefs
1102. ICML 2026What Makes Effective Supervision in Latent Chain-of-Thought: An Information-Theoretic Analysis
1103. ICML 2026What Really Improves Mathematical Reasoning: Structured Reasoning Signals Beyond Pure Code
1104. ICML 2026When Agents Go Rogue: Activation-Based Detection of Malicious Behaviors in Multi-Agent Systems
1105. ICML 2026When Diffusion Language Models Hesitate: Detecting and Correcting Visual Hallucinations via Confidence Fluctuation
1106. ICML 2026When Distance Distracts: Representation Distance Bias in BT-Loss for Reward Models
1107. ICML 2026When Do Hallucinations Arise? A Graph Perspective on the Evolution of Path Reuse and Path Compression
1108. ICML 2026When Does Sparsity Mitigate the Curse of Depth in LLMs
1109. ICML 2026When Drafts Evolve: Speculative Decoding Meets Online Learning
1110. ICML 2026When Labelers Stay Silent: The Power of Ties in Cost-Effective Preference Learning
1111. ICML 2026When LLMs Encounter Open-world Graph Learning: A Fresh View on Unlabeled Data Uncertainty
1112. ICML 2026When Model Merging Breaks Routing: Training-Free Calibration for MoE
1113. ICML 2026When RAG Hurts: Diagnosing and Mitigating Attention Distraction in Retrieval-Augmented LVLMs
1114. ICML 2026When RL Meets Adaptive Speculative Training: A Unified Training-Serving System
1115. ICML 2026When Search Goes Wrong: Red-Teaming Web-Augmented Large Language Models
1116. ICML 2026When Simple Problems Wear Complex Costumes: Improving Efficiency in LRM's Adaptive Reasoning
1117. ICML 2026When to Think, When to Speak: Learning Disclosure Policies for LLM Reasoning
1118. ICML 2026Which Heads Matter for Reasoning? RL-Guided KV Cache Compression
1119. ICML 2026Who can we trust? LLM-as-a-jury for Comparative Assessment
1120. ICML 2026Why Deep Jacobian Spectra Separate: Depth-Induced Scaling and Singular-Vector Alignment
1121. ICML 2026WinDeskGround: A Benchmark for Robust GUI Grounding in Complex Multi-Window Desktop Environments
1122. ICML 2026WinQ: Accelerating Quantization-Aware Training of Language Models Around Saddle Points
1123. ICML 2026Words & Weights: Streamlining Multi-Turn Interactions via Co-Adaptation
1124. ICML 2026WorldTravel: A Realistic Multimodal Travel-Planning Benchmark with Tightly Coupled Constraints
1125. ICML 2026WUSH: Near-Optimal Adaptive Transforms for LLM Quantization
1126. ICML 2026xKV: Cross-Layer KV-Cache Compression via Aligned Singular Vector Extraction
1127. ICML 2026XSkill: Continual Learning from Experience and Skills in Multimodal Agents
1128. ICML 2026XSpecMesh: Quality-Preserving Auto-Regressive Mesh Generation Acceleration via Multi-Head Speculative Decoding
1129. ICML 2026You Can Learn Tokenization End-to-End with Reinforcement Learning
1130. ICML 2026Zero Sum SVD: Balancing Loss Sensitivity for Low Rank LLM Compression
1131. ICML 2026ZipMoE: Efficient On-Device MoE Serving via Lossless Compression and Cache-Affinity Scheduling

常见问题

2023–2026 年四大会议一共有多少篇深度学习 · large language models论文？

共 2,334 篇，按年份分别是 2023 年 488 篇、2024 年 298 篇、2025 年 417 篇、2026 年 1,131 篇。

NeurIPS 2024 有多少篇深度学习 · large language models论文？

NeurIPS 2024 有 298 篇深度学习 · large language models论文，上一届是 415 篇。

ICLR 2025 有多少篇深度学习 · large language models论文？

ICLR 2025 有 13 篇深度学习 · large language models论文。

ICML 2026 有多少篇深度学习 · large language models论文？

ICML 2026 有 1,131 篇深度学习 · large language models论文，上一届是 404 篇。

哪个会议的深度学习 · large language models论文最多？

ICML，2023–2026 年共 1,608 篇。