arXiv subjects

Xiaohua Jia

Publications and source records attributed to Xiaohua Jia.

At least 37 records · Page 2Linked to original sources

arXiv2025

The Hidden Dimensions of LLM Alignment: A Multi-Dimensional Analysis of Orthogonal Safety Directions

Large Language Models' safety-aligned behaviors, such as refusing harmful queries, can be represented by linear directions in activation space. Previous research modeled safety behavior with a single direction, limiting mechanistic understanding to an isolated safety feature. In this work, we discover that safety-aligned behavior is jointly controlled by multi-dimensional directions. Namely, we study the vector space of representation shifts during safety fine-tuning on Llama 3 8B for refusing jailbreaks. By studying orthogonal directions in the space, we first find that a dominant direction governs the model's refusal behavior, while multiple smaller directions represent distinct and interpretable features like hypothetical narrative and role-playing. We then measure how different directions promote or suppress the dominant direction, showing the important role of secondary directions in shaping the model's refusal representation. Finally, we demonstrate that removing certain trigger tokens in harmful queries can mitigate these directions to bypass the learned safety capability, providing new insights on understanding safety alignment vulnerability from a multi-dimensional perspective. Code and artifacts are available at https://github.com/BMPixel/safety-residual-space.

cs.CL↗

arXiv2025

Metadata-private Messaging without Coordination

For those seeking end-to-end private communication free from pervasive metadata tracking and censorship, the Tor network has been the de-facto choice in practice, despite its susceptibility to traffic analysis attacks. Recently, numerous metadata-private messaging proposals have emerged with the aim to surpass Tor in the messaging context by obscuring the relationships between any two messaging buddies, even against global and active attackers. However, most of these systems face an undesirable usability constraint: they require a metadata-private "dialing" phase to establish mutual agreement and timing or round coordination before initiating any regular chats among users. This phase is not only resource-intensive but also inflexible, limiting users' ability to manage multiple concurrent conversations seamlessly. For stringent privacy requirement, the often-enforced traffic uniformity further exacerbated the limitations of this roadblock. In this paper, we introduce PingPong, a new end-to-end system for metadata-private messaging designed to overcome these limitations. Under the same traffic uniformity requirement, PingPong replaces the rigid "dial-before-converse" paradigm with a more flexible "notify-before-retrieval" workflow. This workflow incorporates a metadata-private notification subsystem, Ping, and a metadata-private message store, Pong. Both Ping and Pong leverage hardware-assisted secure enclaves for performance and operates through a series of customized oblivious algorithms, while meeting the uniformity requirements for metadata protection. By allowing users to switch between conversations on demand, PingPong achieves a level of usability akin to modern instant messaging systems, while also offering improved performance and bandwidth utilization for goodput. We have built a prototype of PingPong with 32 8-core servers equipped with enclaves to validate our claims.

cs.CR↗

arXiv2025

Training with Differential Privacy: A Gradient-Preserving Noise Reduction Approach with Provable Security

Deep learning models have been extensively adopted in various regions due to their ability to represent hierarchical features, which highly rely on the training set and procedures. Thus, protecting the training process and deep learning algorithms is paramount in privacy preservation. Although Differential Privacy (DP) as a powerful cryptographic primitive has achieved satisfying results in deep learning training, the existing schemes still fall short in preserving model utility, i.e., they either invoke a high noise scale or inevitably harm the original gradients. To address the above issues, in this paper, we present a more robust and provably secure approach for differentially private training called GReDP. Specifically, we compute the model gradients in the frequency domain and adopt a new approach to reduce the noise level. Unlike previous work, our GReDP only requires half of the noise scale compared to DPSGD [1] while keeping all the gradient information intact. We present a detailed analysis of our method both theoretically and empirically. The experimental results show that our GReDP works consistently better than the baselines on all models and training settings.

cs.CR↗

arXiv2025

Exploring Incremental Unlearning: Techniques, Challenges, and Future Directions

The growing demand for data privacy in Machine Learning (ML) applications has seen Machine Unlearning (MU) emerge as a critical area of research. As the `right to be forgotten' becomes regulated globally, it is increasingly important to develop mechanisms that delete user data from AI systems while maintaining performance and scalability of these systems. Incremental Unlearning (IU) is a promising MU solution to address the challenges of efficiently removing specific data from ML models without the need for expensive and time-consuming full retraining. This paper presents the various techniques and approaches to IU. It explores the challenges faced in designing and implementing IU mechanisms. Datasets and metrics for evaluating the performance of unlearning techniques are discussed as well. Finally, potential solutions to the IU challenges alongside future research directions are offered. This survey provides valuable insights for researchers and practitioners seeking to understand the current landscape of IU and its potential for enhancing privacy-preserving intelligent systems.

cs.LG↗

arXiv2025

LiveVal: Time-aware Data Valuation via Adaptive Reference Points

Time-aware data valuation enhances training efficiency and model robustness, as early detection of harmful samples could prevent months of wasted computation. However, existing methods rely on model retraining or convergence assumptions or fail to capture long-term training dynamics. We propose LiveVal, an efficient time-aware data valuation method with three key designs: 1) seamless integration with SGD training for efficient data contribution monitoring; 2) reference-based valuation with normalization for reliable benchmark establishment; and 3) adaptive reference point selection for real-time updating with optimized memory usage. We establish theoretical guarantees for LiveVal's stability and prove that its valuations are bounded and directionally aligned with optimization progress. Extensive experiments demonstrate that LiveVal provides efficient data valuation across different modalities and model scales, achieving 180 speedup over traditional methods while maintaining robust detection performance.

cs.LG↗

arXiv2024

Protective Perturbations against Unauthorized Data Usage in Diffusion-based Image Generation

Diffusion-based text-to-image models have shown immense potential for various image-related tasks. However, despite their prominence and popularity, customizing these models using unauthorized data also brings serious privacy and intellectual property issues. Existing methods introduce protective perturbations based on adversarial attacks, which are applied to the customization samples. In this systematization of knowledge, we present a comprehensive survey of protective perturbation methods designed to prevent unauthorized data usage in diffusion-based image generation. We establish the threat model and categorize the downstream tasks relevant to these methods, providing a detailed analysis of their designs. We also propose a completed evaluation framework for these perturbation techniques, aiming to advance research in this field.

cs.CV↗

arXiv2024

Embedding Watermarks in Diffusion Process for Model Intellectual Property Protection

In practical application, the widespread deployment of diffusion models often necessitates substantial investment in training. As diffusion models find increasingly diverse applications, concerns about potential misuse highlight the imperative for robust intellectual property protection. Current protection strategies either employ backdoor-based methods, integrating a watermark task as a simpler training objective with the main model task, or embedding watermarks directly into the final output samples. However, the former approach is fragile compared to existing backdoor defense techniques, while the latter fundamentally alters the expected output. In this work, we introduce a novel watermarking framework by embedding the watermark into the whole diffusion process, and theoretically ensure that our final output samples contain no additional information. Furthermore, we utilize statistical algorithms to verify the watermark from internally generated model samples without necessitating triggers as conditions. Detailed theoretical analysis and experimental validation demonstrate the effectiveness of our proposed method.

cs.CV↗

arXiv2024

LMEraser: Large Model Unlearning through Adaptive Prompt Tuning

To address the growing demand for privacy protection in machine learning, we propose a novel and efficient machine unlearning approach for \textbf{L}arge \textbf{M}odels, called \textbf{LM}Eraser. Existing unlearning research suffers from entangled training data and complex model architectures, incurring extremely high computational costs for large models. LMEraser takes a divide-and-conquer strategy with a prompt tuning architecture to isolate data influence. The training dataset is partitioned into public and private datasets. Public data are used to train the backbone of the model. Private data are adaptively clustered based on their diversity, and each cluster is used to optimize a prompt separately. This adaptive prompt tuning mechanism reduces unlearning costs and maintains model performance. Experiments demonstrate that LMEraser achieves a $100$-fold reduction in unlearning costs without compromising accuracy compared to prior work. Our code is available at: \url{https://github.com/lmeraser/lmeraser}.

cs.LG↗

arXiv2024

Machine Unlearning: Solutions and Challenges

Machine learning models may inadvertently memorize sensitive, unauthorized, or malicious data, posing risks of privacy breaches, security vulnerabilities, and performance degradation. To address these issues, machine unlearning has emerged as a critical technique to selectively remove specific training data points' influence on trained models. This paper provides a comprehensive taxonomy and analysis of the solutions in machine unlearning. We categorize existing solutions into exact unlearning approaches that remove data influence thoroughly and approximate unlearning approaches that efficiently minimize data influence. By comprehensively reviewing solutions, we identify and discuss their strengths and limitations. Furthermore, we propose future directions to advance machine unlearning and establish it as an essential capability for trustworthy and adaptive machine learning models. This paper provides researchers with a roadmap of open problems, encouraging impactful contributions to address real-world needs for selective data removal.

cs.LG↗

arXiv2023

Intellectual Property Protection of Diffusion Models via the Watermark Diffusion Process

Diffusion models have rapidly become a vital part of deep generative architectures, given today's increasing demands. Obtaining large, high-performance diffusion models demands significant resources, highlighting their importance as intellectual property worth protecting. However, existing watermarking techniques for ownership verification are insufficient when applied to diffusion models. Very recent research in watermarking diffusion models either exposes watermarks during task generation, which harms the imperceptibility, or is developed for conditional diffusion models that require prompts to trigger the watermark. This paper introduces WDM, a novel watermarking solution for diffusion models without imprinting the watermark during task generation. It involves training a model to concurrently learn a Watermark Diffusion Process (WDP) for embedding watermarks alongside the standard diffusion process for task generation. We provide a detailed theoretical analysis of WDP training and sampling, relating it to a shifted Gaussian diffusion process via the same reverse noise. Extensive experiments are conducted to validate the effectiveness and robustness of our approach in various trigger and watermark data configurations.

cs.CR↗

arXiv2023

On the Security Bootstrapping in Named Data Networking

By requiring all data packets been cryptographically authenticatable, the Named Data Networking (NDN) architecture design provides a basic building block for secured networking. This basic NDN function requires that all entities in an NDN network go through a security bootstrapping process to obtain the initial security credentials. Recent years have witnessed a number of proposed solutions for NDN security bootstrapping protocols. Built upon the existing results, in this paper we take the next step to develop a systematic model of security bootstrapping: Trust-domain Entity Bootstrapping (TEB). This model is based on the emerging concept of trust domain and describes the steps and their dependencies in the bootstrapping process. We evaluate the expressiveness and sufficiency of this model by using it to describe several current bootstrapping protocols.

cs.CR↗

arXiv2023

SecGNN: Privacy-Preserving Graph Neural Network Training and Inference as a Cloud Service

Graphs are widely used to model the complex relationships among entities. As a powerful tool for graph analytics, graph neural networks (GNNs) have recently gained wide attention due to its end-to-end processing capabilities. With the proliferation of cloud computing, it is increasingly popular to deploy the services of complex and resource-intensive model training and inference in the cloud due to its prominent benefits. However, GNN training and inference services, if deployed in the cloud, will raise critical privacy concerns about the information-rich and proprietary graph data (and the resulting model). While there has been some work on secure neural network training and inference, they all focus on convolutional neural networks handling images and text rather than complex graph data with rich structural information. In this paper, we design, implement, and evaluate SecGNN, the first system supporting privacy-preserving GNN training and inference services in the cloud. SecGNN is built from a synergy of insights on lightweight cryptography and machine learning techniques. We deeply examine the procedure of GNN training and inference, and devise a series of corresponding secure customized protocols to support the holistic computation. Extensive experiments demonstrate that SecGNN achieves comparable plaintext training and inference accuracy, with promising performance.

cs.CR↗

arXiv2022

SecSkyline: Fast Privacy-Preserving Skyline Queries over Encrypted Cloud Databases

The well-known benefits of cloud computing have spurred the popularity of database service outsourcing, where one can resort to the cloud to conveniently store and query databases. Coming with such popular trend is the threat to data privacy, as the cloud gains access to the databases and queries which may contain sensitive information, like medical or financial data. A large body of work has been presented for querying encrypted databases, which has been mostly focused on secure keyword search. In this paper, we instead focus on the support for secure skyline query processing over encrypted outsourced databases, where little work has been done. Skyline query is an advanced kind of database query which is important for multi-criteria decision-making systems and applications. We propose SecSkyline, a new system framework building on lightweight cryptography for fast privacy-preserving skyline queries. SecSkyline ambitiously provides strong protection for not only the content confidentiality of the outsourced database, the query, and the result, but also for data patterns that may incur indirect data leakages, such as dominance relationships among data points and search access patterns. Extensive experiments demonstrate that SecSkyline is substantially superior to the state-of-the-art in query latency, with up to 813$\times$ improvement.

cs.CR↗

arXiv2022

OblivGM: Oblivious Attributed Subgraph Matching as a Cloud Service

In recent years there has been growing popularity of leveraging cloud computing for storing and querying attributed graphs, which have been widely used to model complex structured data in various applications. Such trend of outsourced graph analytics, however, is accompanied with critical privacy concerns regarding the information-rich and proprietary attributed graph data. In light of this, we design, implement, and evaluate OblivGM, a new system aimed at oblivious graph analytics services outsourced to the cloud. OblivGM focuses on the support for attributed subgraph matching, one popular and fundamental graph query functionality aiming to retrieve from a large attributed graph subgraphs isomorphic to a small query graph. Built from a delicate synergy of insights from attributed graph modelling and advanced lightweight cryptography, OblivGM protects the confidentiality of data content associated with attributed graphs and queries, conceals the connections among vertices in attributed graphs, and hides search access patterns. Meanwhile, OblivGM flexibly supports oblivious evaluation of varying subgraph queries, which may contain equality and/or range predicates. Extensive experiments over a real-world attributed graph dataset demonstrate that while providing strong security guarantees, OblivGM achieves practically affordable performance (with query latency on the order of a few seconds).

cs.CR↗

arXiv2022

Privacy-Preserving Analytics on Decentralized Social Graphs: The Case of Eigendecomposition

Analytics over social graphs allows to extract valuable knowledge and insights for many fields like community detection, fraud detection, and interest mining. In practice, decentralized social graphs frequently arise, where the social graph is not available to a single entity and is decentralized among a large number of users, each holding only a limited local view about the whole graph. Collecting the local views for analytics of decentralized social graphs raises critical privacy concerns, as they encode private information about the social interactions among individuals. In this paper, we design, implement, and evaluate PrivGED, a new system aimed at privacy-preserving analytics over decentralized social graphs. PrivGED focuses on the support for eigendecomposition, one popular and fundamental graph analytics task producing eigenvalues/eigenvectors over the adjacency matrix of a social graph and benefits various practical applications. PrivGED is built from a delicate synergy of insights on graph analytics, lightweight cryptography, and differential privacy, allowing users to securely contribute their local views on a decentralized social graph for a cloud-based eigendecomposition analytics service while gaining strong privacy protection. Extensive experiments over real-world social graph datasets demonstrate that PrivGED achieves accuracy comparable to the plaintext domain, with practically affordable performance superior to prior art.

cs.CR↗

arXiv2021

Detecting and Identifying Optical Signal Attacks on Autonomous Driving Systems

For autonomous driving, an essential task is to detect surrounding objects accurately. To this end, most existing systems use optical devices, including cameras and light detection and ranging (LiDAR) sensors, to collect environment data in real time. In recent years, many researchers have developed advanced machine learning models to detect surrounding objects. Nevertheless, the aforementioned optical devices are vulnerable to optical signal attacks, which could compromise the accuracy of object detection. To address this critical issue, we propose a framework to detect and identify sensors that are under attack. Specifically, we first develop a new technique to detect attacks on a system that consists of three sensors. Our main idea is to: 1) use data from three sensors to obtain two versions of depth maps (i.e., disparity) and 2) detect attacks by analyzing the distribution of disparity errors. In our study, we use real data sets and the state-of-the-art machine learning model to evaluate our attack detection scheme and the results confirm the effectiveness of our detection method. Based on the detection scheme, we further develop an identification model that is capable of identifying up to n-2 attacked sensors in a system with one LiDAR and n cameras. We prove the correctness of our identification scheme and conduct experiments to show the accuracy of our identification method. Finally, we investigate the overall sensitivity of our framework.

cs.CV↗

arXiv2021

Evaluating Adversarial Attacks on Driving Safety in Vision-Based Autonomous Vehicles

In recent years, many deep learning models have been adopted in autonomous driving. At the same time, these models introduce new vulnerabilities that may compromise the safety of autonomous vehicles. Specifically, recent studies have demonstrated that adversarial attacks can cause a significant decline in detection precision of deep learning-based 3D object detection models. Although driving safety is the ultimate concern for autonomous driving, there is no comprehensive study on the linkage between the performance of deep learning models and the driving safety of autonomous vehicles under adversarial attacks. In this paper, we investigate the impact of two primary types of adversarial attacks, perturbation attacks and patch attacks, on the driving safety of vision-based autonomous vehicles rather than the detection precision of deep learning models. In particular, we consider two state-of-the-art models in vision-based 3D object detection, Stereo R-CNN and DSGN. To evaluate driving safety, we propose an end-to-end evaluation framework with a set of driving safety performance metrics. By analyzing the results of our extensive evaluation experiments, we find that (1) the attack's impact on the driving safety of autonomous vehicles and the attack's impact on the precision of 3D object detectors are decoupled, and (2) the DSGN model demonstrates stronger robustness to adversarial attacks than the Stereo R-CNN model. In addition, we further investigate the causes behind the two findings with an ablation study. The findings of this paper provide a new perspective to evaluate adversarial attacks and guide the selection of deep learning models in autonomous driving.

cs.CV↗

arXiv2021

Throughput Maximization of UAV Networks

In this paper we study the deployment of multiple unmanned aerial vehicles (UAVs) to form a temporal UAV network for the provisioning of emergent communications to affected people in a disaster zone, where each UAV is equipped with a lightweight base station device and thus can act as an aerial base station for users. Unlike most existing studies that assumed that a UAV can serve all users in its communication range, we observe that both computation and communication capabilities of a single lightweight UAV are very limited, due to various constraints on its size, weight, and power supply. Thus, a single UAV can only provide communication services to a limited number of users. We study a novel problem of deploying $K$ UAVs in the top of a disaster area such that the sum of the data rates of users served by the UAVs is maximized, subject to that (i) the number of users served by each UAV is no greater than its service capacity; and (ii) the communication network induced by the $K$ UAVs is connected. We then propose a $\frac{1-1/e}{\lfloor \sqrt{K} \rfloor}$-approximation algorithm for the problem, improving the current best result of the problem by five times (the best approximation ratio so far is $\frac{1-1/e}{5( \sqrt{K} +1)}$), where $e$ is the base of the natural logarithm. We finally evaluate the algorithm performance via simulation experiments. Experimental results show that the proposed algorithm is very promising. Especially, the solution delivered by the proposed algorithm is up to 12% better than those by existing algorithms.

cs.DC↗