arXiv subjects

Weixian Li

Publications and source records attributed to Weixian Li.

4 recordsLinked to original sources

arXiv2025

DC-SGD: Differentially Private SGD with Dynamic Clipping through Gradient Norm Distribution Estimation

Differentially Private Stochastic Gradient Descent (DP-SGD) is a widely adopted technique for privacy-preserving deep learning. A critical challenge in DP-SGD is selecting the optimal clipping threshold C, which involves balancing the trade-off between clipping bias and noise magnitude, incurring substantial privacy and computing overhead during hyperparameter tuning. In this paper, we propose Dynamic Clipping DP-SGD (DC-SGD), a framework that leverages differentially private histograms to estimate gradient norm distributions and dynamically adjust the clipping threshold C. Our framework includes two novel mechanisms: DC-SGD-P and DC-SGD-E. DC-SGD-P adjusts the clipping threshold based on a percentile of gradient norms, while DC-SGD-E minimizes the expected squared error of gradients to optimize C. These dynamic adjustments significantly reduce the burden of hyperparameter tuning C. The extensive experiments on various deep learning tasks, including image classification and natural language processing, show that our proposed dynamic algorithms achieve up to 9 times acceleration on hyperparameter tuning than DP-SGD. And DC-SGD-E can achieve an accuracy improvement of 10.62% on CIFAR10 than DP-SGD under the same privacy budget of hyperparameter tuning. We conduct rigorous theoretical privacy and convergence analyses, showing that our methods seamlessly integrate with the Adam optimizer. Our results highlight the robust performance and efficiency of DC-SGD, offering a practical solution for differentially private deep learning with reduced computational overhead and enhanced privacy guarantees.

cs.LG↗

arXiv2025

R.R.: Unveiling LLM Training Privacy through Recollection and Ranking

Large Language Models (LLMs) pose significant privacy risks, potentially leaking training data due to implicit memorization. Existing privacy attacks primarily focus on membership inference attacks (MIAs) or data extraction attacks, but reconstructing specific personally identifiable information (PII) in LLMs' training data remains challenging. In this paper, we propose R.R. (Recollect and Rank), a novel two-step privacy stealing attack that enables attackers to reconstruct PII entities from scrubbed training data where the PII entities have been masked. In the first stage, we introduce a prompt paradigm named recollection, which instructs the LLM to repeat a masked text but fill in masks. Then we can use PII identifiers to extract recollected PII candidates. In the second stage, we design a new criterion to score each PII candidate and rank them. Motivated by membership inference, we leverage the reference model as a calibration to our criterion. Experiments across three popular PII datasets demonstrate that the R.R. achieves better PII identification performance than baselines. These results highlight the vulnerability of LLMs to PII leakage even when training data has been scrubbed. We release our code and datasets at GitHub.

cs.CL↗

arXiv2023

Weight Distribution of Repeated-Root Cyclic Codes with Prime Power Lengths

Determining the weight distribution of a linear code is a classical and fundamental topic in coding theory that has been extensively investigated. Repeated-root cyclic codes, which form a significant subclass of error-correcting codes, have found broad applications in quantum error-correcting codes, symbol-pair codes, and storage codes. Through polynomial derivation, we derive the monomial equivalent codes for these repeated-root cyclic codes with prime power lengths. Given that monomial equivalent codes exhibit identical weight distributions, we transform the computation of the weight distribution of these repeated-root cyclic codes into the computation of the weight distribution of their monomial equivalent codes. Leveraging the classical results on the weight distribution of MDS codes, we explicitly determine the weight distribution of these repeated-root cyclic codes. Moreover, we apply the weight distribution formula to construct a class of $p$-weight cyclic codes for any prime $p$.

cs.IT↗

arXiv2021

A Characterization of MDS Symbol-pair Codes over Two Types of Alphabets

Symbol-pair codes are block codes with symbol-pair metrics designed to protect against pair-errors that may occur in high-density data storage systems. MDS symbol-pair codes are optimal in the sense that it can attain the highest pair-error correctability within the same code length and code size. Constructing MDS symbol-pair codes is one of the main topics in symbol-pair codes. In this paper, we characterize the symbol-pair distances of some constacyclic codes of arbitrary lengths over finite fields and a class of finite chain rings. Using the characterization of symbol-pair distance, we present several classes of MDS symbol-pair constacyclic codes and show that there is no other MDS symbol-pair code among the class of constacyclic codes except for what we present. Moreover, some of these MDS symbol-pair constacyclic codes over the finite chain rings cannot be obtained by previous work.

cs.IT↗