5 papers
Few Tokens, Big Leverage: Preserving Safety Alignment by Constraining Safety Tokens during Fine-tuning
Guoli Wang, Haonan Shi, Tu Ouyang +1
Large language models (LLMs) often require fine-tuning (FT) to perform well on downstream tasks, but FT can induce safety-alignment drift even when the training dataset contains on…
EASE: Practical and Efficient Safety Alignment for Small Language Models
Haonan Shi, Guoli Wang, Tu Ouyang +1
Small language models (SLMs) are increasingly deployed on edge devices, making their safety alignment crucial yet challenging. Current shallow alignment methods that rely on direct…
Unveiling Client Privacy Leakage from Public Dataset Usage in Federated Distillation
Haonan Shi, Tu Ouyang, An Wang
Federated Distillation (FD) has emerged as a popular federated training framework, enabling clients to collaboratively train models without sharing private data. Public Dataset-Ass…
Navigating the Designs of Privacy-Preserving Fine-tuning for Large Language Models
Haonan Shi, Tu Ouyang, An Wang
Instruction tuning has proven effective in enhancing Large Language Models' (LLMs) performance on downstream tasks. However, real-world fine-tuning faces inherent conflicts between…
Learning-Based Difficulty Calibration for Enhanced Membership Inference Attacks
Haonan Shi, Tu Ouyang, An Wang
Machine learning models, in particular deep neural networks, are currently an integral part of various applications, from healthcare to finance. However, using sensitive data to tr…