Privacy Accounting for Empirical Denoisers
Diffusion models have emerged as the foundation of modern generative systems, yet their high memorization capacity raise
#stat.ML
共收录 38 条相关情报。
← 返回所有主题Reliable Estimation of Differential Privacy under Selection Bias
Statistical privacy audits are essential for verifying the theoretical privacy guarantees of differentially private (DP)
Private Evaluation of High-Dimensional Functions
We study differentially private black-box evaluation of arbitrary high-dimensional functions. Given a dataset $S$ and qu
Task-aware Kernel Stein certificates for Rényi differential privacy
Rènyi differential privacy, based on Rènyi divergence, has become a notion of choice for privacy guarantees. However, ou
Towards Hyperparameter Transfer for Differentially Private Optimization
Scaling laws and hyperparameter tuning remain central challenges in training large-scale models, particularly under diff
Efficient Agnostic Learning of Univariate Densities Under Local Differential Privacy
We propose a general framework for non-interactive locally differentially private (LDP) learning of structured univariat
Limits of Statistical Assurance for Embodied and Agentic AI under Adaptive Adversaries
Finite safety assessments cannot exclude an adversary that preserves unattacked behavior until certification and activat
Tight End-to-End Accounting for Differentially Private Synthetic Data
Differentially private (DP) pipelines often combine heterogeneous mechanisms, yet their guarantees are best analyzed usi
Can Rényi Accountants Beat Lossless Accountants for DP-SGD?
Recent frameworks in differential privacy (DP), such as privacy loss distributions and $f$-DP, provide lossless represen
Accuracy and Uncertainty in Machine Deception Detection (Research White Paper)
A deception detector is only as useful as the trust its alarms deserve. Like a polygraph, detectors that read a language
Verification with Transfer: Exact Information Frontiers and Their Price in Calls
A verifier that accepts or rejects whole answers reveals little: under a flat prior over $k$-bit answers, zero error nee
One-Shot Private Confidence Regions via Resampling
We propose a simple framework for constructing differentially private confidence regions \textit{in one shot}, i.e., by
Private online learning and prediction for Littlestone classes
We study mistake bounds for differentially private online learning and online prediction under oblivious realisable adve
Watermarking: from Impossibility to Auditable Compliance
Article 50 (2) of the EU Artificial Intelligence Act requires providers of generative systems to make synthetic outputs
Differential Privacy of Gradient Descent on Perturbed Objectives
本文研究目标扰动（objective perturbation）机制在有限步优化下的差分隐私性质。经典目标扰动做法是在正则化经验风险上叠加一个随机线性项 ⟨z,w⟩（其中 z 从 N(0, σ²I_d) 中一次性采样），随后发布扰动目标的精
Understanding Private Evolution as Learning-Augmented Clustering
本文研究差分隐私合成数据生成算法 Private Evolution（PE）的理论性质与改进方案。PE 是一种在差分隐私约束下生成合成数据的迭代式算法，其工作方式可以形式化为一种 Wasserstein（最优传输）学习过程：算法反复从生成器
Retrospective Distillation Attribution via Normalized Response Similarity
模型蒸馏通过在监督微调（SFT）阶段使用教师模型（通常来自商业 API）生成的回复来迁移能力，这使蒸馏后模型的来源归属（provenance）成为合规、许可与知识产权层面的现实问题。已有的蒸馏归因方法大多假设学生模型在 SFT 刚结束时就被
The Like Trap: Multi-Stage Poisoning against Agents in Similarity-based Recommendation Systems
随着大语言模型及其驱动的智能体（agent）能力增强，这些智能体日益自主化，并可代表用户在互联网上执行操作，例如自动管理用户的社交媒体账号。然而，部署在社交平台上的自动化智能体自身的脆弱性研究仍不充分。现有关于智能体投毒的研究通常假设攻击者
Benchmarking Quantum Machine Learning for Power-System Attack Detection: Evaluation Choices Decide the Outcome Before the Models Do
本论文针对电力系统网络攻击检测中的机器学习模型鲁棒性问题，系统评估了量子机器学习（QML）方案与传统经典模型在对抗环境下的表现，并提出一个关键方法论论点：评估者自身的选择会先于模型决定基准测试的结论。研究在公开的密西西比州立大学/ORNL
Context Is Not Authority: Structured Runtime Governance for Financial Market Agents
本文提出 SAGE-Fin，一个面向金融场景智能体（agent）的运行时治理框架，核心论点是“上下文不等于授权”。作者指出，金融智能体在对话或任务执行中可能将正确的上下文错误地转化为未经授权的实际动作，例如做出客户承诺、执行交易或部署策略。