Bowen Deng

Differential privacy (DP) mechanisms have been widely adopted in federated learning (FL) to enhance model security. However, existing approaches predominantly employ uniform privacy budgets, neglecting personalized requirements arising from heterogeneous user privacy preferences. Such uniform privacy configurations typically necessitate compliance with the most stringent budget, which not only leads to the wasteful underutilization of privacy budgets for certain clients but also compromises overall model utility. To address this limitation, we propose FedSPA, a Subspace Projection Aggregation personalized differential private Federated learning framework. The proposed method conducts singular value decomposition operations on noise-perturbed local models to extract singular value vectors as compact representations of both model structure and privacy noise. The server then clusters clients and identifies a consensus subspace for projecting models with varying noise levels, ultimately aggregating the global model through a residual-aware mechanism. This method not only effectively guides the aggregation of client personalized differential privacy but also reduces communication overhead. Extensive experiments demonstrate the model's effectiveness. Additionally, we provide theoretical proof of the privacy and convergence of FedSPA. Experimental results also showcase its superior performance over personalized DP-FL baselines.

Current adversarial defense methods for GNNs exhibit critical limitations obstructing real-world application: 1) inadequate adaptability to graph heterophily, 2) absent generalizability to early GNNs like GraphSAGE used downstream, and 3) low inference scalability unacceptable for resource-constrained scenarios. To simultaneously address these challenges, we propose the first online GNN-MLP distillation framework PROSPECT, which merges the complementary knowledge of MLP and GNN and can thus learn GNN and MLP robust against adversarial structure attacks on both homophilic and heterophilic graphs. PROSPECT integrates seamlessly into GraphSAGE and achieves inference scalability exponentially higher than conventional GNNs. To mitigate potential convergence failure caused by inductive bias conflicts between the heterogeneous MLP and GNN, we propose the Quasi-Alternating Cosine Annealing (QACA) learning rate scheduler, inspired by our convergence analysis of the involved MLP. Experiments on homophilic and heterophilic graphs demonstrate the advantages of PROSPECT over current defenses and offline GNN-MLP distillation methods in terms of adversarial robustness and clean accuracy, the inference scalability of PROSPECT orders of magnitude higher than existing defenses, and the effectiveness of QACA.

Accurate estimation of time difference of arrival (TDOA) between sensors is a fundamental problem in various applications of sensor networks. In this letter, a novel algorithm is proposed to estimate TDOAs based on low-rank essence of a TDOA matrix, whose entry (i, j) denotes TDOA value between sensors i and j with respect to a common reference source. To enhance the robustness of the proposed algorithm, we further introduce the Gram matrix of the TDOA matrix, resulting in a semidefinite programming estimation problem. An efficient alternating direction method of multipliers is developed to improve the scalability of the proposed algorithm. Simulation results validate the performance of the proposed algorithm.