代码生成论文趋势 2023–2026

主要结论

• 代码生成论文从 2023 年的 210 篇增长到 2026 年的 860 篇，三年间扩大到约 4.1 倍。
• NeurIPS 是发代码生成论文最多的会议，2023–2026 年共 630 篇，占四个会议总量的 29%。
• 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 93 篇。
• 在「自然语言」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增强生成是 5.4 倍，代码生成是 4.1 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 Theoretical Analysis of Test-Driven Code Generation
• ICML 2026$A_2$DEPT: Large Language Model–Driven Automated Algorithm Design via Evolutionary Program Trees
• ICLR 2026A Bayesian Nonparametric Framework for Private, Fair, and Balanced Tabular Data Synthesis
• CVPR 2026ACPV-Net All-Class Polygonal Vectorization for Seamless Vector Map Generation from
• NeurIPS 2025$\texttt{STRCMP}$: Integrating Graph Structural Priors with Language Models for Combinatorial Optimization
• ICML 2025AlphaVerus: Bootstrapping Formally Verified Code Generation through Self-Improving Translation and Treefinement
• ICLR 2025Automated Proof Generation for Rust Code via Self-Evolution
• CVPR 2025ArtFormer: Controllable Generation of Diverse 3D Articulated Objects
• NeurIPS 2024AMBROSIA: A Benchmark for Parsing Ambiguous Questions into Database Queries
• ICML 2024ACM-MILP: Adaptive Constraint Modification via Grouping and Selection for Hardness-Preserving MILP Instance Generation
• ICLR 2024CODE REPRESENTATION LEARNING AT SCALE
• CVPR 2024Desigen: A Pipeline for Controllable Design Template Generation
• NeurIPS 2023ANPL: Towards Natural Programming with Interactive Decomposition
• ICML 2023Can Large Language Models Reason about Program Invariants?
• ICLR 2023Binding Language Models in Symbolic Languages
• CVPR 2023Attribute-Preserving Face Dataset Anonymization via Latent Code Optimization
• NeurIPS 2026A Theory of Time-Sensitive Language Generation: Sparse Hallucination Beats Mode Collapse
• ICML 2026Beyond Problem Solving: UOJ-Bench for Evaluating Code Generation, Hacking, and Repair in Competitive Programming
• ICLR 2026AssetFormer: Modular 3D Assets Generation with Autoregressive Transformer
• CVPR 2026Animator-Centric Skeleton Generation on Objects with Fine-Grained Details

全部代码生成论文

共 79 篇，第 2 / 2 页

51. ICML 2025NextCoder: Robust Adaptation of Code LMs to Diverse Code Edits
52. ICML 2025NICE Data Selection for Instruction Tuning in LLMs with Non-differentiable Evaluation Metric
53. ICML 2025On Exact Bit-level Reversible Transformers Without Changing Architecture
54. ICML 2025On the Query Complexity of Verifier-Assisted Language Generation
55. ICML 2025Optimizing Language Models for Inference Time Objectives using Reinforcement Learning
56. ICML 2025Overcoming Non-monotonicity in Transducer-based Streaming Generation
57. ICML 2025polybasic Speculative Decoding Through a Theoretical Perspective
58. ICML 2025Position: Future Research and Challenges Remain Towards AI for Software Engineering
59. ICML 2025Prior Knowledge Guided Neural Architecture Generation
60. ICML 2025ProSec: Fortifying Code LLMs with Proactive Security Alignment
61. ICML 2025QuEST: Stable Training of LLMs with 1-Bit Weights and Activations
62. ICML 2025Reasoning Through Execution: Unifying Process and Outcome Rewards for Code Generation
63. ICML 2025Reasoning-as-Logic-Units: Scaling Test-Time Reasoning in Large Language Models Through Logic Unit Alignment
64. ICML 2025Representative Language Generation
65. ICML 2025Revisiting Chain-of-Thought in Code Generation: Do Language Models Need to Learn Reasoning before Coding?
66. ICML 2025Scalable Non-Equivariant 3D Molecule Generation via Rotational Alignment
67. ICML 2025Selective Prompt Anchoring for Code Generation
68. ICML 2025Self-Improving Language Models for Evolutionary Program Synthesis: A Case Study on ARC-AGI
69. ICML 2025SongGen: A Single Stage Auto-regressive Transformer for Text-to-Song Generation
70. ICML 2025SpeCache: Speculative Key-Value Caching for Efficient Generation of LLMs
71. ICML 2025Strategic Planning: A Top-Down Approach to Option Generation
72. ICML 2025Synthesizing Software Engineering Data in a Test-Driven Manner
73. ICML 2025The Diffusion Duality
74. ICML 2025Thinking LLMs: General Instruction Following with Thought Generation
75. ICML 2025TokenSwift: Lossless Acceleration of Ultra Long Sequence Generation
76. ICML 2025Towards Practical Defect-Focused Automated Code Review
77. ICML 2025Tractable Transformers for Flexible Conditional Generation
78. ICML 2025TransPL: VQ-Code Transition Matrices for Pseudo-Labeling of Time Series Unsupervised Domain Adaptation
79. ICML 2025Understanding Complexity in VideoQA via Visual Program Generation

常见问题

2023–2026 年四大会议一共有多少篇代码生成论文？

共 2,161 篇，按年份分别是 2023 年 210 篇、2024 年 446 篇、2025 年 645 篇、2026 年 860 篇。

NeurIPS 2026 有多少篇代码生成论文？

NeurIPS 2026 有 236 篇代码生成论文，上一届是 230 篇。

ICLR 2026 有多少篇代码生成论文？

ICLR 2026 有 241 篇代码生成论文，上一届是 164 篇。

ICML 2026 有多少篇代码生成论文？

ICML 2026 有 174 篇代码生成论文，上一届是 79 篇。

CVPR 2026 有多少篇代码生成论文？

CVPR 2026 有 209 篇代码生成论文，上一届是 172 篇。

哪个会议的代码生成论文最多？

NeurIPS，2023–2026 年共 630 篇。