隐私攻击论文趋势 2023–2026

主要结论

• 隐私攻击论文从 2023 年的 124 篇增长到 2026 年的 219 篇，三年间扩大到约 1.8 倍。
• ICLR 是发隐私攻击论文最多的会议，2023–2026 年共 262 篇，占四个会议总量的 38%。
• 交叉方向里涨得最快的是对齐与安全：同时属于两个主题的论文从 2023 年的 2 篇增至 2026 年的 80 篇。
• 在「联邦与隐私」的 5 个主题里，它的 2026 年论文数排第 1；同组增长最快的安全聚合是 1.8 倍，隐私攻击是 1.8 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 Private Empirical Defense Against Privacy Audits
• ICML 2026A Statistical Framework for Analyzing Specification Resistance to Learnware-Inversion Risks
• ICLR 2026(Token-Level) \textbf{InfoRMIA}: Stronger Membership Inference and Privacy Assessment for LLMs
• CVPR 2026Out of Sight Out of Track Adversarial Attacks on Propagation-based
• NeurIPS 2025A Reliable Cryptographic Framework for Empirical Machine Unlearning Evaluation
• ICML 2025A Certified Unlearning Approach without Access to Source Data
• ICLR 2025Adversarial Attacks on Data Attribution
• CVPR 2025Adv-CPG: A Customized Portrait Generation Framework with Facial Adversarial Attacks
• NeurIPS 2024Reproducibility Study on Adversarial Attacks Against Robust Transformer Trackers
• ICML 2024Collective Certified Robustness against Graph Injection Attacks
• ICLR 2024Be Careful What You Smooth For: Label Smoothing Can Be a Privacy Shield but Also a Catalyst for Model Inversion Attacks
• CVPR 2024Efficient Privacy-Preserving Visual Localization Using 3D Ray Clouds
• NeurIPS 2023Bounding the Invertibility of Privacy-preserving Instance Encoding using Fisher Information
• ICML 2023Data Poisoning Attacks Against Multimodal Encoders
• ICLR 2023Adversarial Attacks on Adversarial Bandits
• CVPR 2023Can't Steal? Cont-Steal! Contrastive Stealing Attacks Against Image Encoders
• NeurIPS 2026Adaptive Target-Charging with Privacy Filters and Individual Accounting
• ICML 2026How does Bayesian Sampling help Membership Inference Attacks?
• ICLR 2026Beyond Membership: Limitations of Add/Remove Adjacency in Differential Privacy
• CVPR 2026Revisiting Geometric Obfuscation with Dual Convergent Lines for Privacy-Preserving Image

全部隐私攻击论文

共 29 篇，第 1 / 1 页

1. ICML 2025A Certified Unlearning Approach without Access to Source Data
2. ICML 2025A Closer Look at Backdoor Attacks on CLIP
3. ICML 2025Auditing Prompt Caching in Language Model APIs
4. ICML 2025Black-Box Adversarial Attacks on LLM-Based Code Completion
5. ICML 2025Connecting Thompson Sampling and UCB: Towards More Efficient Trade-offs Between Privacy and Regret
6. ICML 2025De-AntiFake: Rethinking the Protective Perturbations Against Voice Cloning Attacks
7. ICML 2025Differential Privacy Guarantees of Markov Chain Monte Carlo Algorithms
8. ICML 2025EncryptedLLM: Privacy-Preserving Large Language Model Inference via GPU-Accelerated Fully Homomorphic Encryption
9. ICML 2025Flexible, Efficient, and Stable Adversarial Attacks on Machine Unlearning
10. ICML 2025Fragments to Facts: Partial-Information Fragment Inference from LLMs
11. ICML 2025HyperNear: Unnoticeable Node Injection Attacks on Hypergraph Neural Networks
12. ICML 2025ICLShield: Exploring and Mitigating In-Context Learning Backdoor Attacks
13. ICML 2025Improving the Variance of Differentially Private Randomized Experiments through Clustering
14. ICML 2025Interpreting the Repeated Token Phenomenon in Large Language Models
15. ICML 2025Language Models May Verbatim Complete Text They Were Not Explicitly Trained On
16. ICML 2025Leveraging Per-Instance Privacy for Machine Unlearning
17. ICML 2025MELON: Provable Defense Against Indirect Prompt Injection Attacks in AI Agents
18. ICML 2025On Differential Privacy for Adaptively Solving Search Problems via Sketching
19. ICML 2025Position: Contextual Integrity is Inadequately Applied to Language Models
20. ICML 2025Privacy Attacks on Image AutoRegressive Models
21. ICML 2025Private Lossless Multiple Release
22. ICML 2025Revealing Weaknesses in Text Watermarking Through Self-Information Rewrite Attacks
23. ICML 2025Textual Unlearning Gives a False Sense of Unlearning
24. ICML 2025The Canary’s Echo: Auditing Privacy Risks of LLM-Generated Synthetic Text
25. ICML 2025The Ripple Effect: On Unforeseen Complications of Backdoor Attacks
26. ICML 2025The Role of Randomness in Stability
27. ICML 2025Theoretically Unmasking Inference Attacks Against LDP-Protected Clients in Federated Vision Models
28. ICML 2025Towards LLM Unlearning Resilient to Relearning Attacks: A Sharpness-Aware Minimization Perspective and Beyond
29. ICML 2025Underestimated Privacy Risks for Minority Populations in Large Language Model Unlearning

常见问题

2023–2026 年四大会议一共有多少篇隐私攻击论文？

共 698 篇，按年份分别是 2023 年 124 篇、2024 年 180 篇、2025 年 175 篇、2026 年 219 篇。

NeurIPS 2026 有多少篇隐私攻击论文？

NeurIPS 2026 有 55 篇隐私攻击论文，上一届是 45 篇。

ICLR 2026 有多少篇隐私攻击论文？

ICLR 2026 有 102 篇隐私攻击论文，上一届是 83 篇。

ICML 2026 有多少篇隐私攻击论文？

ICML 2026 有 38 篇隐私攻击论文，上一届是 29 篇。

CVPR 2026 有多少篇隐私攻击论文？

CVPR 2026 有 24 篇隐私攻击论文，上一届是 18 篇。

哪个会议的隐私攻击论文最多？

ICLR，2023–2026 年共 262 篇。