5 papers
Private Direct Preference Optimization for LLM Alignment
Yangfan Jiang, Fei Wei, Ergute Bao +3
Direct preference optimization (DPO) is now a standard method for aligning large language models (LLMs) using human preference data. Each DPO example contains a prompt and a pair o…
Auditing Apple's DifferentialPrivacy.framework: Implementation Bugs, Misconfigurations, and Practical Risks
Rishav Chourasia, Ergute Bao, Uzair Javaid +1
Since 2016, Apple has claimed that device analytics collected to improve user experience are protected by differential privacy (DP). Apple's DifferentialPrivacy framework is deploy…
Overcoming the Retrieval Barrier: Indirect Prompt Injection in the Wild for LLM Systems
Hongyan Chang, Ergute Bao, Xinjian Luo +1
Large language models (LLMs) increasingly rely on retrieving information from external corpora. This creates a new attack surface: indirect prompt injection (IPI), where hidden ins…
Accurate Table Question Answering with Accessible LLMs
Yangfan Jiang, Fei Wei, Ergute Bao +4
Given a table T in a database and a question Q in natural language, the table question answering (TQA) task aims to return an accurate answer to Q based on the content of T. Recent…
GCON: Differentially Private Graph Convolutional Network via Objective Perturbation
Jianxin Wei, Yizheng Zhu, Xiaokui Xiao +4
Graph Convolutional Networks (GCNs) are a popular machine learning model with a wide range of applications in graph analytics, including healthcare, transportation, and finance. Ho…