鲁棒基准论文趋势 2023–2026

主要结论

• 鲁棒基准论文从 2023 年的 64 篇增长到 2026 年的 69 篇，三年间扩大到约 1.1 倍。
• NeurIPS 是发鲁棒基准论文最多的会议，2023–2026 年共 80 篇，占四个会议总量的 30%。
• 交叉方向里涨得最快的是通用机器学习 · evaluation：同时属于两个主题的论文从 2023 年的 0 篇增至 2026 年的 8 篇。
• 在「评测与基准」的 5 个主题里，它的 2026 年论文数排第 4；同组增长最快的大模型基准是 154 倍，鲁棒基准是 1.1 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 2026Brittlebench: Quantifying LLM robustness via prompt sensitivity
• ICML 2026Geometrically Constrained Outlier Synthesis
• ICLR 2026Pitfalls in Evaluating Language Model Forecasters
• CVPR 2026FoV-Net Rotation-Invariant CAD B-rep Learning via Field-of-View Ray Casting
• NeurIPS 2025Aggregation Hides Out-of-Distribution Generalization Failures from Spurious Correlations
• ICML 2025Diverse Prototypical Ensembles Improve Robustness to Subpopulation Shift
• ICLR 2025DRoP: Distributionally Robust Data Pruning
• CVPR 2025FRAMES-VQA: Benchmarking Fine-Tuning Robustness across Multi-Modal Shifts in Visual Question Answering
• NeurIPS 2024An Analysis of Robustness of Non-Lipschitz Networks
• ICML 2024Controllable Prompt Tuning For Balancing Group Distributional Robustness
• ICLR 2024Realistic Evaluation of Semi-supervised Learning Algorithms in Open Environments
• CVPR 2024A Bayesian Approach to OOD Robustness in Image Classification
• NeurIPS 2023Revisiting Out-of-distribution Robustness in NLP: Benchmarks, Analysis, and LLMs Evaluations
• ICML 2023A Law of Robustness beyond Isoperimetry
• ICLR 2023A Call to Reflect on Evaluation Practices for Failure Detection in Image Classification
• CVPR 2023Improving Robustness of Vision Transformers by Reducing Sensitivity To Patch Corruptions
• NeurIPS 2026Stress-Testing Neural Network Verifiers with Provably Robust Instances
• ICML 2026Optimal Attention Temperature Improves the Robustness of In-Context Learning under Distribution Shift in High Dimensions
• ICLR 2026CLARC: C/C++ Benchmark for Robust Code Search
• CVPR 2026LIBERO-Plus A Progressive Robustness Benchmark for Visual-Language-Action Models

全部鲁棒基准论文

共 20 篇，第 1 / 1 页

1. ICML 2025Adversarial Robustness via Deformable Convolution with Stochasticity
2. ICML 2025Diverse Prototypical Ensembles Improve Robustness to Subpopulation Shift
3. ICML 2025Explaining the role of Intrinsic Dimensionality in Adversarial Training
4. ICML 2025LAION-C: An Out-of-Distribution Benchmark for Web-Scale Vision Models
5. ICML 2025LEVIS: Large Exact Verifiable Input Spaces for Neural Networks
6. ICML 2025Mahalanobis++: Improving OOD Detection via Feature Normalization
7. ICML 2025Off-Policy Actor-Critic for Adversarial Observation Robustness: Virtual Alternative Training via Symmetric Policy Evaluation
8. ICML 2025One Stone, Two Birds: Enhancing Adversarial Defense Through the Lens of Distributional Discrepancy
9. ICML 2025Pixel-level Certified Explanations via Randomized Smoothing
10. ICML 2025Position: AI Safety Must Embrace an Antifragile Perspective
11. ICML 2025Position: AI Should Not Be An Imitation Game: Centaur Evaluations
12. ICML 2025Position: Certified Robustness Does Not (Yet) Imply Model Security
13. ICML 2025Position: Graph Learning Will Lose Relevance Due To Poor Benchmarks
14. ICML 2025Position: Medical Large Language Model Benchmarks Should Prioritize Construct Validity
15. ICML 2025Position: Societal Impacts Research Requires Benchmarks for Creative Composition Tasks
16. ICML 2025Probably Approximately Global Robustness Certification
17. ICML 2025Quadratic Upper Bound for Boosting Robustness
18. ICML 2025Statistical Hypothesis Testing for Auditing Robustness in Language Models
19. ICML 2025Towards Robust Influence Functions with Flat Validation Minima
20. ICML 2025When and How Does CLIP Enable Domain and Compositional Generalization?

常见问题

2023–2026 年四大会议一共有多少篇鲁棒基准论文？

共 264 篇，按年份分别是 2023 年 64 篇、2024 年 59 篇、2025 年 72 篇、2026 年 69 篇。

NeurIPS 2026 有多少篇鲁棒基准论文？

NeurIPS 2026 有 18 篇鲁棒基准论文，上一届是 34 篇。

ICLR 2026 有多少篇鲁棒基准论文？

ICLR 2026 有 15 篇鲁棒基准论文，上一届是 10 篇。

ICML 2026 有多少篇鲁棒基准论文？

ICML 2026 有 24 篇鲁棒基准论文，上一届是 20 篇。

CVPR 2026 有多少篇鲁棒基准论文？

CVPR 2026 有 12 篇鲁棒基准论文，上一届是 8 篇。

哪个会议的鲁棒基准论文最多？

NeurIPS，2023–2026 年共 80 篇。