Zhigang Lü

Zhigang Lu, Hassan Jameel Asghar, Mohamed Ali Kaafar et al.

Differential privacy (DP) has been applied in deep learning for preserving privacy of the underlying training sets. Existing DP practice falls into three categories - objective perturbation, gradient perturbation and output perturbation. They suffer from three main problems. First, conditions on objective functions limit objective perturbation in general deep learning tasks. Second, gradient perturbation does not achieve a satisfactory privacy-utility trade-off due to over-injected noise in each epoch. Third, high utility of the output perturbation method is not guaranteed because of the loose upper bound on the global sensitivity of the trained model parameters as the noise scale parameter. To address these problems, we analyse a tighter upper bound on the global sensitivity of the model parameters. Under a black-box setting, based on this global sensitivity, to control the overall noise injection, we propose a novel output perturbation framework by injecting DP noise into a randomly sampled neuron (via the exponential mechanism) at the output layer of a baseline non-private neural network trained with a convexified loss function. We empirically compare the privacy-utility trade-off, measured by accuracy loss to baseline non-private models and the privacy leakage against black-box membership inference (MI) attacks, between our framework and the open-source differentially private stochastic gradient descent (DP-SGD) approaches on six commonly used real-world datasets. The experimental evaluations show that, when the baseline models have observable privacy leakage under MI attacks, our framework achieves a better privacy-utility trade-off than existing DP-SGD implementations, given an overall privacy budget $ε\leq 1$ for a large number of queries.

Shuchao Pang, Zhigang Lu, Haichen Wang et al.

Differential privacy (DP) is the de facto privacy standard against privacy leakage attacks, including many recently discovered ones against large language models (LLMs). However, we discovered that LLMs could reconstruct the altered/removed privacy from given DP-sanitized prompts. We propose two attacks (black-box and white-box) based on the accessibility to LLMs and show that LLMs could connect the pair of DP-sanitized text and the corresponding private training data of LLMs by giving sample text pairs as instructions (in the black-box attacks) or fine-tuning data (in the white-box attacks). To illustrate our findings, we conduct comprehensive experiments on modern LLMs (e.g., LLaMA-2, LLaMA-3, ChatGPT-3.5, ChatGPT-4, ChatGPT-4o, Claude-3, Claude-3.5, OPT, GPT-Neo, GPT-J, Gemma-2, and Pythia) using commonly used datasets (such as WikiMIA, Pile-CC, and Pile-Wiki) against both word-level and sentence-level DP. The experimental results show promising recovery rates, e.g., the black-box attacks against the word-level DP over WikiMIA dataset gave 72.18% on LLaMA-2 (70B), 82.39% on LLaMA-3 (70B), 75.35% on Gemma-2, 91.2% on ChatGPT-4o, and 94.01% on Claude-3.5 (Sonnet). More urgently, this study indicates that these well-known LLMs have emerged as a new security risk for existing DP text sanitization approaches in the current environment.

Aoting Hu, Zhigang Lu, Renjie Xie et al.

Deploying Machine Learning as a Service gives rise to model plagiarism, leading to copyright infringement. Ownership testing techniques are designed to identify model fingerprints for verifying plagiarism. However, previous works often rely on overfitting or robustness features as fingerprints, lacking theoretical guarantees and exhibiting under-performance on generalized models. In this paper, we propose a novel ownership testing method called VeriDIP, which verifies a DNN model's intellectual property. VeriDIP makes two major contributions. (1) It utilizes membership inference attacks to estimate the lower bound of privacy leakage, which reflects the fingerprint of a given model. The privacy leakage fingerprints highlight the unique patterns through which the models memorize sensitive training datasets. (2) We introduce a novel approach using less private samples to enhance the performance of ownership testing. Extensive experimental results confirm that VeriDIP is effective and efficient in validating the ownership of deep learning models trained on both image and tabular datasets. VeriDIP achieves comparable performance to state-of-the-art methods on image datasets while significantly reducing computation and communication costs. Enhanced VeriDIP demonstrates superior verification performance on generalized deep learning models, particularly on table-trained models. Additionally, VeriDIP exhibits similar effectiveness on utility-preserving differentially private models compared to non-differentially private baselines.

Aoting Hu, Renjie Xie, Zhigang Lu et al.

Generative Adversarial Networks (GAN)-synthesized table publishing lets people privately learn insights without access to the private table. However, existing studies on Membership Inference (MI) Attacks show promising results on disclosing membership of training datasets of GAN-synthesized tables. Different from those works focusing on discovering membership of a given data point, in this paper, we propose a novel Membership Collision Attack against GANs (TableGAN-MCA), which allows an adversary given only synthetic entries randomly sampled from a black-box generator to recover partial GAN training data. Namely, a GAN-synthesized table immune to state-of-the-art MI attacks is vulnerable to the TableGAN-MCA. The success of TableGAN-MCA is boosted by an observation that GAN-synthesized tables potentially collide with the training data of the generator. Our experimental evaluations on TableGAN-MCA have five main findings. First, TableGAN-MCA has a satisfying training data recovery rate on three commonly used real-world datasets against four generative models. Second, factors, including the size of GAN training data, GAN training epochs and the number of synthetic samples available to the adversary, are positively correlated to the success of TableGAN-MCA. Third, highly frequent data points have high risks of being recovered by TableGAN-MCA. Fourth, some unique data are exposed to unexpected high recovery risks in TableGAN-MCA, which may attribute to GAN's generalization. Fifth, as expected, differential privacy, without the consideration of the correlations between features, does not show commendable mitigation effect against the TableGAN-MCA. Finally, we propose two mitigation methods and show promising privacy and utility trade-offs when protecting against TableGAN-MCA.

Zhigang Lu, Hong Shen

Iterative clustering algorithms help us to learn the insights behind the data. Unfortunately, this may allow adversaries to infer the privacy of individuals with some background knowledge. In the worst case, the adversaries know the centroids of an arbitrary iteration and the information of n-1 out of n items. To protect individual privacy against such an inference attack, preserving differential privacy (DP) for the iterative clustering algorithms has been extensively studied in the interactive settings. However, existing interactive differentially private clustering algorithms suffer from a non-convergence problem, i.e., these algorithms may not terminate without a predefined number of iterations. This problem severely impacts the clustering quality and the efficiency of a differentially private algorithm. To resolve this problem, in this paper, we propose a novel differentially private clustering framework in the interactive settings which controls the orientation of the movement of the centroids over the iterations to ensure the convergence by injecting DP noise in a selected area. We prove that, in the expected case, algorithm under our framework converges in at most twice the iterations of Lloyd's algorithm. We perform experimental evaluations on real-world datasets to show that our algorithm outperforms the state-of-the-art of the interactive differentially private clustering algorithms with guaranteed convergence and better clustering quality to meet the same DP requirement.

Zhigang Lu, Hong Shen

Paths in a given network are a generalised form of time-serial chains in many real-world applications, such as trajectories and Internet flows. Differentially private trajectory publishing concerns publishing path information that is usable to the genuine users yet secure against adversaries to reconstruct the path with maximum background knowledge. The exiting studies all assume this knowledge to be all but one vertex on the path. To prevent the adversaries recovering the missing information, they publish a perturbed path where each vertex is sampled from a pre-defined set with differential privacy (DP) to replace the corresponding vertex in the original path. In this paper, we relax this assumption to be all but one edge on the path, and hence consider the scenario of more powerful adversaries with the maximum background knowledge of the entire network topology and the path (including all the vertices) except one (arbitrary) missing edge. Under such an assumption, the perturbed path produced by the existing work is vulnerable, because the adversary can reconstruct the missing edge from the existence of an edge in the perturbed path. To address this vulnerability and effectively protect edge privacy, instead of publishing a perturbed path, we propose a novel scheme of graph-based path publishing to protect the original path by embedding the path in a graph that contains fake edges and replicated vertices applying the differential privacy technique, such that only the legitimate users who have the full knowledge of the network topology are able to recover the exact vertices and edges of the original path with high probability. We theoretically analyse the performance of our algorithm in differential privacy, utility, and execution efficiency. We also conduct extensive experimental evaluations on a high-performance cluster system to validate our analytical results.

Zhigang Lu, Hong Shen

The neighbourhood-based Collaborative Filtering is a widely used method in recommender systems. However, the risks of revealing customers' privacy during the process of filtering have attracted noticeable public concern recently. Specifically, $k$NN attack discloses the target user's sensitive information by creating $k$ fake nearest neighbours by non-sensitive information. Among the current solutions against $k$NN attack, the probabilistic methods showed a powerful privacy preserving effect. However, the existing probabilistic methods neither guarantee enough prediction accuracy due to the global randomness, nor provide assured security enforcement against $k$NN attack. To overcome the problems of current probabilistic methods, we propose a novel approach, Partitioned Probabilistic Neighbour Selection, to ensure a required security guarantee while achieving the optimal prediction accuracy against $k$NN attack. In this paper, we define the sum of $k$ neighbours' similarity as the accuracy metric $α$, the number of user partitions, across which we select the $k$ neighbours, as the security metric $β$. Differing from the present methods that globally selected neighbours, our method selects neighbours from each group with exponential differential privacy to decrease the magnitude of noise. Theoretical and experimental analysis show that to achieve the same security guarantee against $k$NN attack, our approach ensures the optimal prediction accuracy.

Zhigang Lu, Hong Shen

Recommender systems, tool for predicting users' potential preferences by computing history data and users' interests, show an increasing importance in various Internet applications such as online shopping. As a well-known recommendation method, neighbourhood-based collaborative filtering has attracted considerable attention recently. The risk of revealing users' private information during the process of filtering has attracted noticeable research interests. Among the current solutions, the probabilistic techniques have shown a powerful privacy preserving effect. When facing $k$ Nearest Neighbour attack, all the existing methods provide no data utility guarantee, for the introduction of global randomness. In this paper, to overcome the problem of recommendation accuracy loss, we propose a novel approach, Partitioned Probabilistic Neighbour Selection, to ensure a required prediction accuracy while maintaining high security against $k$NN attack. We define the sum of $k$ neighbours' similarity as the accuracy metric alpha, the number of user partitions, across which we select the $k$ neighbours, as the security metric beta. We generalise the $k$ Nearest Neighbour attack to beta k Nearest Neighbours attack. Differing from the existing approach that selects neighbours across the entire candidate list randomly, our method selects neighbours from each exclusive partition of size $k$ with a decreasing probability. Theoretical and experimental analysis show that to provide an accuracy-assured recommendation, our Partitioned Probabilistic Neighbour Selection method yields a better trade-off between the recommendation accuracy and system security.