Federal Learning (FL) is highly respected for protecting data privacy in a distributed environment. However, the correlation between the updated gradient and the training data opens up the possibility of data reconstruction for malicious attackers, thus threatening the basic privacy requirements of FL. Previous research on such attacks mainly focuses on two main perspectives: one exclusively relies on gradient attacks, which performs well on small-scale data but falter with large-scale data; the other incorporates images prior but faces practical implementation challenges. So far, the effectiveness of privacy leakage attacks in FL is still far from satisfactory. In this paper, we introduce the Gradient Guided Diffusion Model (GGDM), a novel learning-free approach based on a pre-trained unconditional Denoising Diffusion Probabilistic Models (DDPM), aimed at improving the effectiveness and reducing the difficulty of implementing gradient based privacy attacks on complex networks and high-resolution images. To the best of our knowledge, this is the first work to employ the DDPM for privacy leakage attacks of FL. GGDM capitalizes on the unique nature of gradients and guides DDPM to ensure that reconstructed images closely mirror the original data. In addition, in GGDM, we elegantly combine the gradient similarity function with the Stochastic Differential Equation (SDE) to guide the DDPM sampling process based on theoretical analysis, and further reveal the impact of common similarity functions on data reconstruction. Extensive evaluation results demonstrate the excellent generalization ability of GGDM. Specifically, compared with state-of-the-art methods, GGDM shows clear superiority in both quantitative metrics and visualization, significantly enhancing the reconstruction quality of privacy attacks.

The rapid growth of online social networks has created significant opportunities for large-scale information dissemination, where many users seek to maximize the visibility and influence of their information, bringing significant attention to the problem of influence maximization (IM). IM aims to identify a limited set of influential users (seed nodes) to maximize information spread. However, existing IM approaches typically provide a unified solution for all users under a fixed seed budget, without considering that different users in real-world scenarios have inherently different target diffusion ranges and seed budgets. Such omissions lead to inefficiency and resource waste, particularly when excessive seed budgets are allocated to users with low diffusion demands. To overcome this issue, we introduce Personalized Influence Maximization (PIM) as an extension of classical IM. Building upon this formulation, we propose the Adaptive Graph Influencer Generator (AGIG), which models seed set selection as a sequence generation task and employs a causal transformer to autoregressively generate personalized and cost-effective seed sets tailored to diverse demands of users. In particular, AGIG incorporates an enhanced dual-view influence encoder that models realistic scenarios where information can reach users without direct connections but with similar interests, thereby strengthening node representations for high-quality seed generation. For effective training, we construct the Propagation Pathways Sequence Dataset by simulating diffusion processes under multiple classical diffusion models and graph structures, enabling AGIG to learn diverse propagation patterns across varying diffusion settings. Extensive experiments demonstrate that AGIG effectively adapts to diverse personalized propagation requirements, achieving an average improvement of approximately 15% in influence spread and cost efficiency over strong baseline methods across multiple datasets and diffusion settings.