2023安全顶会AI安全相关汇总

NDSS

对抗样本防御

1. Adversarial Robustness for Tabular Data through Cost and Utility Awareness
2. BARS: Local Robustness Certification for Deep Learning based Traffic Analysis Systems 流量相关

后门攻击

1. Backdoor Attacks Against Dataset Distillation

S&P

AI伦理

1. How technical do you get? I’m an English teacher”: Teaching and Learning Cybersecurity and AI Ethics in High School

AI&差分隐私

1. A Theory to Instruct Differentially-Private Learning via Clipping Bias Reduction
2. Continual Observation under User-level Differential Privacy
3. Locally Differentially Private Frequency Estimation Based on Convolution Framework
4. Spectral-DP: Differentially Private Deep Learning through Spectral Perturbation and Filtering

对抗样本攻击

1. AI-Guardian: Defeating Adversarial Attacks using Backdoors
2. SoK: Certified Robustness for Deep Neural Networks

后门攻击

1. Redeem Myself: Purifying Backdoors in Deep Learning Models using Self Attention Distillation
2. Disguising Attacks with Explanation-Aware Backdoors

推理攻击

1. SNAP: Efficient Extraction of Private Properties with Poisoning
2. Accuracy-Privacy Trade-off in Deep Ensemble: A Membership Inference Perspective
3. SoK: Let the Privacy Games Begin! A Unified Treatment of Data Inference Privacy in Machine Learning

模型萃取攻击

1. D-DAE: Defense-Penetrating Model Extraction Attacks

机器学习可解释性

1. Everybody’s Got ML, Tell Me What Else You Have: Practitioners’ Perception of ML-Based Security Tools and Explanations

USENIX

对抗样本攻击

1. KENKU: Towards Efficient and Stealthy Black-box Adversarial Attacks against ASR Systems
2. Towards Targeted Obfuscation of Adversarial Unsafe Images using Reconstruction and Counterfactual Super Region Attribution Explainability
3. TPatch: A Triggered Physical Adversarial Patch
4. CAPatch: Physical Adversarial Patch against Image Captioning Systems
5. Hard-label Black-box Universal Adversarial Patch Attack
6. The Space of Adversarial Strategies
7. X-Adv: Physical Adversarial Object Attacks against X-ray Prohibited Item Detection
8. SMACK: Semantically Meaningful Adversarial Audio Attack audio
9. URET: Universal Robustness Evaluation Toolkit (for Evasion)
10. Precise and Generalized Robustness Certification for Neural Networks
11. DiffSmooth: Certifiably Robust Learning via Diffusion Models and Local Smoothing

成员推理攻击

后门攻击

1. Towards A Proactive ML Approach for Detecting Backdoor Poison Samples
2. PELICAN: Exploiting Backdoors of Naturally Trained Deep Learning Models In Binary Code Analysis 二进制代码分析自动发现后门
3. A Data-free Backdoor Injection Approach in Neural Networks
4. Sparsity Brings Vulnerabilities: Exploring New Metrics in Backdoor Attacks
5. Aliasing Backdoor Attacks on Pre-trained Models
6. ASSET: Robust Backdoor Data Detection Across a Multiplicity of Deep Learning Paradigms
7. VILLAIN: Backdoor Attacks Against Vertical Split Learning
8. FreeEagle: Detecting Complex Neural Trojans in Data-Free Cases

投毒攻击

1. Meta-Sift: How to Sift Out a Clean Subset in the Presence of Data Poisoning?
2. Fine-grained Poisoning Attack to Local Differential Privacy Protocols for Mean and Variance Estimation

比特翻转(bit-flip)攻击

1. Aegis: Mitigating Targeted Bit-flip Attacks against Deep Neural Networks
2. NeuroPots: Realtime Proactive Defense against Bit-Flip Attacks in Neural Networks

差分隐私

1. What Are the Chances? Explaining the Epsilon Parameter in Differential Privacy
2. Tight Auditing of Differentially Private Machine Learning

模型水印

1. Rethinking White-Box Watermarks on Deep Learning Models under Neural Structural Obfuscation

其它

1. CodexLeaks: Privacy Leaks from Code Generation Language Models in GitHub Copilot
2. IvySyn: Automated Vulnerability Discovery in Deep Learning Frameworks
3. “Security is not my field, I’m a stats guy”: A Qualitative Root Cause Analysis of Barriers to Adversarial Machine Learning Defenses in Industry 对抗训练相关

CCS

后门攻击

1. Narcissus: A Practical Clean-Label Backdoor Attack with Limited Information

模型窃取

1. Stealing the Decoding Algorithms of Language Models语言模型的解码算法、超参数窃取
2. Stolen Risks of Models with Security Properties 强化学习模型隐私风险验证

差分隐私&机器学习

1. DPMLBench: Holistic Evaluation of Differentially Private Machine Learning
2. Geometry of Sensitivity: Twice Sampling and Hybrid Clipping in Differential Privacy with Optimal Gaussian Noise and Application to Deep Learning
3. Blink: Link Local Differential Privacy in Graph Neural Networks via Bayesian Estimation 图神经网络
4. DP-Forward: Fine-tuning and Inference on Language Models with Differential Privacy in Forward Pass 语言模型

其它

1. Stateful Defenses for Machine Learning Models Are Not Yet Secure Against Black-box Attacks 黑盒攻击增强策略
2. Prediction Privacy in Distributed Multi-Exit Neural Networks: Vulnerabilities and Solutions
3. Devil in Disguise: Breaching Graph Neural Networks Privacy through Infiltration 对图神经网络的攻击
4. Evading Watermark based Detection of AI-Generated Content 生成模型水印检测规避
5. Interactive Proofs For Differentially Private Counting 交互式差分隐私证明（可能没有AI相关的内容）
6. SalsaPicante: A Machine Learning Attack on LWE with Binary Secrets 用机器学习攻击量子密码系统
7. Efficient Query-Based Attack against ML-Based Android Malware Detection under Zero Knowledge Setting 攻击恶意软件检测模型
8. “Get in Researchers; We’re Measuring Reproducibility”: A Reproducibility Study of Machine Learning Papers in Tier 1 Security Conferences 论文可复现性检查
9. DE-FAKE: Detection and Attribution of Fake Images Generated by Text-to-Image Generation Models 生成图像检测
10. Attack Some while Protecting Others: Selective Attack Strategies for Attacking and Protecting Multiple Concepts
11. Unforgeability in Stochastic Gradient Descent SGD执行的可伪造性