人工评测论文趋势 2023–2026

主要结论

• 人工评测论文从 2023 年的 33 篇增长到 2026 年的 108 篇，三年间扩大到约 3.3 倍。
• NeurIPS 是发人工评测论文最多的会议，2023–2026 年共 90 篇，占四个会议总量的 33%。
• 交叉方向里涨得最快的是数据集与基准：同时属于两个主题的论文从 2023 年的 3 篇增至 2026 年的 31 篇。
• 在「评测与基准」的 5 个主题里，它的 2026 年论文数排第 3；同组增长最快的大模型基准是 154 倍，人工评测是 3.3 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

• NeurIPS 2026Ad-Hoc Teamwork from Human Demonstrations
• ICML 2026SWE-IF: Aligning Code Evaluation with Human Preference
• ICLR 2026Human-LLM Collaborative Feature Engineering for Tabular Data
• CVPR 2026Bridging Human Evaluation to Infrared and Visible Image Fusion
• NeurIPS 2025Deep Edge Filter: Return of the Human-Crafted Layer in Deep Learning
• ICML 2025Position: When Incentives Backfire, Data Stops Being Human
• ICLR 2025Eliciting Human Preferences with Language Models
• CVPR 2025DynPose: Largely Improving the Efficiency of Human Pose Estimation by a Simple Dynamic Framework
• NeurIPS 2024DART-Eval: A Comprehensive DNA Language Model Evaluation Benchmark on Regulatory DNA
• ICML 2024A Fine-grained Analysis of Fitted Q-evaluation: Beyond Parametric Models
• ICLR 2024Human Feedback is not Gold Standard
• CVPR 2024LocLLM: Exploiting Generalizable Human Keypoint Localization via Large Language Model
• NeurIPS 2023Privacy Assessment on Reconstructed Images: Are Existing Evaluation Metrics Faithful to Human Perception?
• ICML 2023Regions of Reliability in the Evaluation of Multivariate Probabilistic Forecasts
• ICLR 2023Beyond calibration: estimating the grouping loss of modern neural networks
• CVPR 20233D Human Keypoints Estimation From Point Clouds in the Wild Without Human Labels
• NeurIPS 2026AI Construct Lexis: An Ontology of the Hidden Assumptions in AI Evaluation
• ICML 2026Rethinking Evaluation Paradigms in IBP-based Certified Training
• ICLR 2026Multi-turn Evaluation of Anthropomorphic Behaviours in Large Language Models
• CVPR 2026Ground Reaction Inertial Poser Physics-based Human Motion Capture from Sparse

全部人工评测论文

共 271 篇，第 2 / 6 页

51. ICML 2026Benchmarking World-Model Learning with Environment-Level Queries
52. ICML 2026Certifying Capabilities from Finite Tests: When Is It Possible?
53. ICML 2026Conversation for Non-verifiable Learning: Self-Evolving Large Language Models through Meta-Evaluation
54. ICML 2026Efficient Inference for Noisy LLM-as-a-Judge Evaluation
55. ICML 2026Expectation Alignment of Language Models for Real-World User Expectations
56. ICML 2026Explanations are a Means to an End: Decision Theoretic Explanation Evaluation
57. ICML 2026FiRE: Fine-grained Ranking Evaluation for Machine Translation
58. ICML 2026GICDM: Mitigating Hubness for Reliable Distance-Based Generative Model Evaluation
59. ICML 2026Hair-Trigger Alignment: Black-Box Evaluation Cannot Guarantee Post-Update Alignment
60. ICML 2026LiveOIBench: Can Large Language Models Outperform Human Contestants in Informatics Olympiads?
61. ICML 2026Moving Out: Physically-grounded Human-AI Collaboration
62. ICML 2026Nonparametric LLM Evaluation from Preference Data
63. ICML 2026PerceptionRubrics: Calibrating Multimodal Evaluation to Human Perception
64. ICML 2026QEDBench: Quantifying the Alignment Gap in Automated Evaluation of University-Level Mathematical Proofs
65. ICML 2026Query-efficient model evaluation using cached responses
66. ICML 2026Rethinking Evaluation Paradigms in IBP-based Certified Training
67. ICML 2026Reward Modeling from Natural Language Human Feedback
68. ICML 2026STABLEVAL: Disagreement-Aware and Stable Evaluation of AI Systems
69. ICML 2026SWE-IF: Aligning Code Evaluation with Human Preference
70. NeurIPS 2026Ad-Hoc Teamwork from Human Demonstrations
71. NeurIPS 2026AI Construct Lexis: An Ontology of the Hidden Assumptions in AI Evaluation
72. NeurIPS 2026AIRA 2: Overcoming Bottlenecks in AI Research Agents
73. NeurIPS 2026AMUSE: Anytime Muon with Stable Gradient Evaluation
74. NeurIPS 2026An Enigma of Artificial Reason: Investigating the Production-Evaluation Gap in Large Reasoning Models
75. NeurIPS 2026BayesJudge: Uncertainty-Aware Bayesian Meta-Evaluation of Human and LLM Judgments
76. NeurIPS 2026BenchRep-T: A Systematic Evaluation of T-Cell Repertoire-Based Disease Diagnostics
77. NeurIPS 2026Beyond Proxy Metrics: MLLM-Based Human Surrogate Evaluation for Explainable AI
78. NeurIPS 2026Correct, Route, Calibrate: Efficient Preference Optimization from Noisy, Heterogeneous Human Feedback
79. NeurIPS 2026Cosine is Human: The Reproducibility Ceiling of Perceptual Similarity
80. NeurIPS 2026Evaluating an Evaluation: Membership Inference Attacks as Machine Unlearning Diagnostics
81. NeurIPS 2026Evaluation of Visual Processing Should Be Human-Centered, Not Metric-Centered
82. NeurIPS 2026From Uncertain Judgments to Calibrated Rankings: Conformal Elo Estimation for LLM Evaluation
83. NeurIPS 2026General Agent Evaluation
84. NeurIPS 2026GT-Free OCR Metrics: A Reference-Free Evaluation Framework for Document OCR Systems
85. NeurIPS 2026Hack-Verifiable Environments: Towards Evaluating Reward Hacking at Scale
86. NeurIPS 2026HearSayBench: Can LLMs Navigate from Abstract Human Rights to Lived Lives?
87. NeurIPS 2026HERO: Improving the Reliability and Sensitivity of Generative Model Evaluation Using Historical Data
88. NeurIPS 2026ImageNet FID is a Pass Check, Not a Finish Line
89. NeurIPS 2026Intra-Option Fitted Q-Evaluation: Evaluating Hierarchical Policies from Non-Hierarchical Data
90. NeurIPS 2026Leak-CURBER: A Leakage-Controlled Multimodal Evaluation Benchmark for Enzymatic Reaction Tasks
91. NeurIPS 2026OdysSim: Building Foundation Models for Human Behavior Simulation
92. NeurIPS 2026Off-Policy Evaluation of Large Language Models via Learned Semantic Bottleneck Embeddings
93. NeurIPS 2026OmniMechanism Design for Human-AI Collaboration with Impressionable Minds
94. NeurIPS 2026OTROPE: Optimal Transport-based Robust Off-policy Evaluation for Large Language Models
95. NeurIPS 2026PhyProbe: Rethinking Physical Consistency Evaluation in Generated Videos
96. NeurIPS 2026Post-Selection Distributional Model Evaluation
97. NeurIPS 2026pyCD: A Unified Benchmark for Reliable Evaluation of Cognitive Diagnosis Models
98. NeurIPS 2026Reporting Practice Matters: The Impact of Reference Choice on Chest X-ray Report Evaluation
99. NeurIPS 2026Rethinking XAI Evaluation: A Human-Centered Audit of Shapley Benchmarks in High-Stakes Settings
100. NeurIPS 2026SALART-VQA: Diagnosing Whether VLMs Understand Salient Artifacts in Generated Images

常见问题

2023–2026 年四大会议一共有多少篇人工评测论文？

共 271 篇，按年份分别是 2023 年 33 篇、2024 年 70 篇、2025 年 60 篇、2026 年 108 篇。

NeurIPS 2026 有多少篇人工评测论文？

NeurIPS 2026 有 39 篇人工评测论文，上一届是 17 篇。

ICLR 2026 有多少篇人工评测论文？

ICLR 2026 有 29 篇人工评测论文，上一届是 16 篇。

ICML 2026 有多少篇人工评测论文？

ICML 2026 有 21 篇人工评测论文，上一届是 14 篇。

CVPR 2026 有多少篇人工评测论文？

CVPR 2026 有 19 篇人工评测论文，上一届是 13 篇。

哪个会议的人工评测论文最多？

NeurIPS，2023–2026 年共 90 篇。