Graph Neural Networks (GNNs) with differential privacy have been proposed to preserve graph privacy when nodes represent personal and sensitive information. However, existing methods ignore that nodes with different importance may yield diverse privacy demands, which may lead to over-protecting some nodes and decrease model utility. In this paper, we study the problem of importance-grained privacy, where nodes contain personal data that need to be kept private but are critical for training a GNN. We propose NAP-GNN, a node-importance-grained privacy-preserving GNN algorithm with privacy guarantees based on adaptive differential privacy to safeguard node information. First, we propose a Topology-based Node Importance Estimation (TNIE) method to infer unknown node importance with neighborhood and centrality awareness. Second, an adaptive private aggregation method is proposed to perturb neighborhood aggregation from node-importance-grain. Third, we propose to privately train a graph learning algorithm on perturbed aggregations in an adaptive residual connection mode over multi-layer convolution for node- wise tasks. The theoretical analysis shows that NAP-GNN can guarantee privacy. Empirical experiments over five real-world graph datasets show that NAP-GNN achieves a better trade-off between privacy and accuracy.

Multimodal Recommendation Systems (MRSs) boost traditional user-item interaction-based methods by incorporating multimodal information. However, existing methods ignore the inherent noise brought by (1) noisy semantic priors in multimodal content, and (2) noisy user interactions in history records, therefore diminishing model performance. To fill this gap, we propose to denoise MRSs by jointly EValuating structure Effectiveness and mitigating Noisy links (EVEN). Firstly, for semantic prior noise in multimodal content, EVEN builds item homogeneous consistency and denoises it by evaluating behavior-driven confidence. Secondly, for noise in user interactions, EVEN updates user feedback by denoising observed interactions following implicit contribution evaluation of high-order representations. Thirdly, EVEN performs cross-modal alignment through self-guided structure learning, reinforcing task-specific inter-modal dependency modeling and cross-modal fusion. Through extensive experiments on three widely-used datasets, EVEN achieves an average improvement of 8.95% and 5.90% in recommendation accuracy compared with LGMRec and FREEDOM, respectively, without extending the total training time.

Pseudo-label learning methods have been widely applied in weakly-supervised temporal action localization. Existing works directly utilize weakly-supervised base model to generate instance-level pseudo-labels for training the fully-supervised detection head. We argue that the noise in pseudo-labels would interfere with the learning of fully-supervised detection head, leading to significant performance leakage. Issues with noisy labels include:(1) inaccurate boundary localization; (2) undetected short action clips; (3) multiple adjacent segments incorrectly detected as one segment. To target these issues, we introduce a two-stage noisy label learning strategy to harness every potential useful signal in noisy labels. First, we propose a frame-level pseudo-label generation model with a context-aware denoising algorithm to refine the boundaries. Second, we introduce an online-revised teacher-student framework with a missing instance compensation module and an ambiguous instance correction module to solve the short-action-missing and many-to-one problems. Besides, we apply a high-quality pseudo-label mining loss in our online-revised teacher-student framework to add different weights to the noisy labels to train more effectively. Our model outperforms the previous state-of-the-art method in detection accuracy and inference speed greatly upon the THUMOS14 and ActivityNet v1.2 benchmarks.

Recent breakthroughs in Multimodal Large Language Models (MLLMs) have gained significant recognition within the deep learning community, where the fusion of the Video Foundation Models (VFMs) and Large Language Models(LLMs) has proven instrumental in constructing robust video understanding systems, effectively surmounting constraints associated with predefined visual tasks. These sophisticated MLLMs exhibit remarkable proficiency in comprehending videos, swiftly attaining unprecedented performance levels across diverse benchmarks. However, their operation demands substantial memory and computational resources, underscoring the continued importance of traditional models in video comprehension tasks. In this paper, we introduce a novel learning paradigm termed MLLM4WTAL. This paradigm harnesses the potential of MLLM to offer temporal action key semantics and complete semantic priors for conventional Weakly-supervised Temporal Action Localization (WTAL) methods. MLLM4WTAL facilitates the enhancement of WTAL by leveraging MLLM guidance. It achieves this by integrating two distinct modules: Key Semantic Matching (KSM) and Complete Semantic Reconstruction (CSR). These modules work in tandem to effectively address prevalent issues like incomplete and over-complete outcomes common in WTAL methods. Rigorous experiments are conducted to validate the efficacy of our proposed approach in augmenting the performance of various heterogeneous WTAL models.

Using extensive training data from SA-1B, the Segment Anything Model (SAM) has demonstrated exceptional generalization and zero-shot capabilities, attracting widespread attention in areas such as medical image segmentation and remote sensing image segmentation. However, its performance in the field of image manipulation detection remains largely unexplored and unconfirmed. There are two main challenges in applying SAM to image manipulation detection: a) reliance on manual prompts, and b) the difficulty of single-view information in supporting cross-dataset generalization. To address these challenges, we develops a cross-view prompt learning paradigm called IMDPrompter based on SAM. Benefiting from the design of automated prompts, IMDPrompter no longer relies on manual guidance, enabling automated detection and localization. Additionally, we propose components such as Cross-view Feature Perception, Optimal Prompt Selection, and Cross-View Prompt Consistency, which facilitate cross-view perceptual learning and guide SAM to generate accurate masks. Extensive experimental results from five datasets (CASIA, Columbia, Coverage, IMD2020, and NIST16) validate the effectiveness of our proposed method.

Weakly supervised temporal action localization aims to learn to locate actions in videos from video-level or point-level labels, avoiding the need for costly frame-level annotations. Unlike previous work that relies solely on visual modality information, we propose incorporating audio information into the weakly supervised temporal action localization task. While audio-visual localization tasks combine audio and visual information for video localization, temporal action localization often deals with action categories that have weak audio cues. To address this, we propose EAV-Mamba, the first audio-visual perception modeling method based on Mamba. Leveraging Mamba’s powerful audio-visual perception capabilities, we developed modules such as Audio-Perceptive Flow Enhancement, Audio-Perceptive RGB Enhancement, and Audio Self-Perceptive Enhancement. Extensive experiments on two publicly available temporal action localization datasets demonstrate that EAV-Mamba achieves efficient audio-visual perception modeling and state-of-the-art performance in weakly supervised temporal action localization tasks.

With increasing cyber attacks over the Internet, network intrusion detection systems (NIDS) have been an indispensable barrier to protecting network security. Taking advantage of automatically capturing topology connections, recent deep graph learning approaches have achieved remarkable performance in distinguishing different types of malicious flows. However, there remain some critical challenges. 1) previous supervised learning methods rely heavily on abundant and high-quality annotated samples, while label annotation requires abundant time and expert knowledge. 2) Centralized methods require all data to be uploaded to a server for learning behavior patterns, which results in high detection latency and critical privacy leakage. 3) Diverse attack scenarios exhibit highly imbalanced distribution, making it hard to characterize abnormal behaviors. To address these issues, we proposed FeCoGraph, a label-aware federated graph contrastive learning framework for intrusion detection in few-shot scenarios. The line graph is introduced to directly process flow embeddings, which are compatible with diverse GNNs. Furthermore, We formulate a graph contrastive learning task to effectively leverage label information, allowing intra-class embeddings more compact than inter-class embeddings. To improve the scalability of NIDS, we utilize federated learning to cover more attack scenarios while protecting data privacy. Experiment results show that FeCoGraph surpass E-graphSAGE with an average 8.36% accuracy on binary classification and 6.77% accuracy on multiclass classification, demonstrating the efficiency of our approach.

Graph Neural Networks (GNNs) have demonstrated remarkable potential in various downstream tasks by effectively capturing the relational dependencies among nodes in graphs. However, this capability also brings significant privacy risks: when GNNs encode topological information and node features into their output, sensitive information can be inadvertently exposed, leading to severe privacy breaches. Existing privacy-preserving GNNs primarily focus on protecting the existence of individual nodes or edges, overlooking practical scenarios where nodes and edges are often publicly accessible and only specific sensitive attributes require protection, resulting in a lack of consideration for attribute sensitivity and challenges in balancing privacy and utility. In this paper, we study the problem of hiding sensitive information during GNN training and limiting its exposure in the outputs, while better defending against attribute inference attacks (AIAs) and achieving improved performance. To achieve this, we propose an attribute-oriented differentially private graph neural network (AODP-GNN) that enforces attribute-specific privacy guarantees through dynamic privacy budgets and relevance-aware noise injection, optimizing the balance between privacy and utility. Specifically, we design a neighborhood-aware private embedding generation mechanism and a mutual information minimization-based optimization strategy that operate before the deep interactions of feature interaction and model optimization to strengthen defense against AIAs. To enhance the balance between privacy and utility, we further develop a relevance-grained noise adaptation technique that dynamically allocates higher noise to less relevant attributes. Theoretical analysis shows that the AODP-GNN satisfies privacy guarantees. Extensive experiments conducted on four real-world datasets demonstrate that our approach can achieve up to around 10.04% and 9.21% higher accuracy compared to the state-of-the-art centrally differentially private GNN ProGAP and DPDGC, and also shows a higher defense capability against AIAs.

Differential-Private Graph Neural Networks (DP-GNNs) have generated remarkable research results, enabling them to effectively tackle the privacy leakage problem in graph learning. However, most DP-GNNs do not consider the temporal-dimensional scenarios. In tasks involving spatiotemporal graph training, such as wireless social networks analysis, the sensitive interactive information in each time graph should be protected. Therefore, we propose spatio-temporal dynamic graph learning with enhanced local differential privacy (LDP-STG). First, we design a weighted graph perturbation encoder based on the Bernoulli distribution and Laplace mechanism, which protects the structures of weighted graphs in time series under edge-level local differential privacy. Second, we employ an attention mechanism to learn dynamic graph node embedding from spatial and temporal dimensions. The theoretical analysis proves that our LDP-STG realizes differential privacy guarantees. We conduct experiments mainly on two communication datasets (i.e., Enron and UCI), which shows that our LDP-STG can achieve better privacy utility tradeoffs compared with traditional mechanisms in terms of snatiotemnoral dimensions.

Blockchain data mining has the potential to reveal the operational status and behavioral patterns of anonymous participants in blockchain systems, thus providing valuable insights into system operation and participant behavior. However, traditional blockchain analysis methods suffer from the problems of being unable to handle the data due to its large volume and complex structure. With powerful computing and analysis capabilities, graph learning can solve the current problems through handling each node's features and linkage relationships separately and exploring the implicit properties of data from a graph perspective. This paper systematically reviews the blockchain data mining tasks based on graph learning approaches. First, we investigate the blockchain data acquisition method, integrate the currently available data analysis tools, and divide the sampling method into rule-based and cluster-based techniques. Second, we classify the graph construction into transaction-based blockchain and account-based methods, and comprehensively analyze the existing blockchain feature extraction methods. Third, we compare the existing graph learning algorithms on blockchain and classify them into traditional machine learning-based, graph representation-based, and graph deep learning-based methods. Finally, we propose future research directions and open issues which are promising to address.

Traffic forecasting is essential in improving and maintaining safety and orderliness in intelligent transportation systems (ITS). As a deep learning approach, graph neural networks (GNN) based spatial-temporal association mining methods are promising in traffic forecasting. However, current GNN-based methods usually require a high number of training data, and when the sample volume is small, the performance of the model drops dramatically. The existing transfer methods can solve this problem by leveraging knowledge from other data-rich areas, but the domain adaption method with access to source data still faces the non-neglectable problem of private information leakage in the source area. A solution that can solve cross-area transfer without access to source data is still missing. In this paper, to fill the gap, we propose a Transferable Federated Inductive Spatial-Temporal Graph Neural Network (T-ISTGNN) framework to transfer spatial-temporal dependency information in cross-area data to accomplish traffic state forecasting. First, we introduce a multi-source model aggregation scheme based on federated learning to retain the traffic information of the source areas. Second, we propose a transfer method between source and target areas based on hypothesis transfer learning to achieve domain adaption under source domain data protection. Third, we propose a GNN-based method called Inductive Spatial-Temporal Graph Neural Network (ISTGNN) for traffic forecasting. Experiments on real-world datasets demonstrate that T-ISTGNN is capable of cross-area traffic state forecasting under the restriction of preserving the privacy of source areas.