The last years have witnessed the emergence of a promising self-supervised learning strategy, referred to as masked autoencoding. However, there is lack theoretical understanding how masking matters on graph autoencoders (GAEs). In this work, we present autoencoder (MaskGAE), framework for graph-structured data. Different from standard GAEs, MaskGAE adopts modeling (MGM) principled pretext task - portion edges and attempting reconstruct missing part with partially visible, unmasked...

Image harmonization is a critical task in computer vision, which aims to adjust the foreground make it compatible with background. Recent works mainly focus on using global transformations (i.e., normalization and color curve rendering) achieve visual consistency. However, these models ignore local consistency their huge model sizes limit ability edge devices. In this paper, we propose hierarchical dynamic network (HDNet) adapt features from view for better feature transformation efficient...

Graph clustering, a fundamental and challenging task in graph mining, aims to classify nodes into several disjoint clusters. In recent years, contrastive learning (GCL) has emerged as dominant line of research clustering advances the new state-of-the-art. However, GCL-based methods heavily rely on augmentations schemes, which may potentially introduce challenges such semantic drift scalability issues. Another promising involves adoption modularity maximization, popular effective measure for...

Anti-fraud machine learning systems are perpetually confronted with the significant challenge of concept drift, driven by continuous and intense evolution fraudulent techniques. That is, outdated models trained on historical behaviors often fall short in addressing evolving tactics malicious users over time. The key issue lies effectively tackling rapid fraudsters' to detect these emerging unforeseen anomalies. In this paper, we propose a solution directly accessing real-time data...

Federated learning (FL) is a distributed machine paradigm that needs collaboration between server and series of clients with decentralized data. To make FL effective in real-world applications, existing work devotes to improving the modeling non-IID In settings, there are intra-client inconsistency comes from imbalanced data modeling, inter-client among heterogeneous client distributions, which not only hinders sufficient representation minority data, but also brings discrepant model...

Semantic segmentation requires pixel-level annotation, which is time-consuming. Active Learning (AL) a promising method for reducing data annotation costs. Due to the gap between aerial and natural images, previous AL methods are not ideal, mainly caused by unreasonable labeling units neglect of class imbalance. Previous based on images or regions, does consider characteristics tasks i.e., network often makes mistakes in edge region, interlaced irregular. Therefore, an edge-guided unit...

Knowledge Graph Embedding (KGE) is a fundamental technique that extracts expressive representation from knowledge graph (KG) to facilitate diverse downstream tasks. The emerging federated KGE (FKGE) collaboratively trains distributed KGs held among clients while avoiding exchanging clients' sensitive raw KGs, which can still suffer privacy threats as evidenced in other model trainings (e.g., neural networks). However, quantifying and defending against such remain unexplored for FKGE...

Recently, collaborative learning is proposed to amortize massive computation costs of highly sophisticated artificial intelligence (AI) tasks. To attract lots participants, researchers investigate blockchains ' economic incentives with proof useful work (PoUW) consensus protocols motivate substantial numbers miners in a mining pool complete AI However, participants might be untrusted and defraud rewards as less possible efforts. In the paper, we propose robust efficient scheme called RPoL...

Graph convolutional networks (GCNs) have been shown to be vulnerable small adversarial perturbations, which becomes a severe threat and largely limits their applications in security-critical scenarios. To mitigate such threat, considerable research efforts devoted increasing the robustness of GCNs against attacks. However, current defense approaches are typically designed prevent from untargeted attacks focus on overall performance, making it challenging protect important local nodes more...

Most state-of-the-art deep domain adaptation techniques align source and target samples in a global fashion. That is, after alignment, each sample is expected to become similar any sample. However, alignment may not always be optimal or necessary practice. For example, consider cross-domain fraud detection, where there are two types of transactions: credit non-credit. Aligning non-credit transactions separately yield better performance than as unlikely exhibit patterns transactions. To...

In the era of deep learning, it is critical to protect intellectual property high-performance neural network (DNN) models. Existing proposals, however, are subject adversarial ownership forgery (e.g., methods based on watermarks or fingerprints) require full access original training dataset for verification requiring replay learning process). this paper, we propose a novel Provenance Training (PoT) scheme, first empirical study towards verifying DNN model without accessing any while being...

Online GUI navigation on mobile devices has driven a lot of attention recent years since it contributes to many real-world applications. With the rapid development large language models (LLM), multimodal (MLLM) have tremendous potential this task. However, existing MLLMs need high quality data improve its abilities making correct decisions according human user inputs. In paper, we developed novel and highly valuable dataset, named \textbf{E-ANT}, as first Chinese dataset that contains real...

Graph clustering, a fundamental and challenging task in graph mining, aims to classify nodes into several disjoint clusters. In recent years, contrastive learning (GCL) has emerged as dominant line of research clustering advances the new state-of-the-art. However, GCL-based methods heavily rely on augmentations schemes, which may potentially introduce challenges such semantic drift scalability issues. Another promising involves adoption modularity maximization, popular effective measure for...

Sequential recommendation (SR) aims to predict the next purchasing item according users' dynamic preference learned from their historical user-item interactions. To improve performance of recommendation, learning heterogeneous cross-type behavior dependencies is indispensable for recommender system. However, there still exists some challenges in Multi-Behavior Recommendation (MBSR). On one hand, existing methods only model multi-behavior at behavior-level or item-level, and modelling...

Data augmentation has been widely studied as it can be used to improve the generalizability of graph representation learning models. However, existing works focus only on data homogeneous graphs. for heterogeneous graphs remains under-explored. Considering that contain different types nodes and links, ignoring type information directly applying methods will lead suboptimal results. In this paper, we propose a novel Multi-Aspect Heterogeneous Graph Augmentation framework named MAHGA....

In many practical binary classification applications, such as financial fraud detection or medical diagnosis, it is crucial to optimize a model's performance on high-confidence samples whose scores are higher than specific threshold, which calculated by given false positive rate according requirements. However, the proportion of typically extremely small, especially in long-tailed datasets, can lead poor recall results and an alignment bias between realistic goals loss. To address this...

Over the past few years, graph neural networks (GNNs) have become powerful and practical tools for learning on (static) graph-structure data. However, many real-world applications, such as social e-commerce, involve temporal graphs where nodes edges are dynamically evolving. Temporal (TGNNs) progressively emerged an extension of GNNs to address time-evolving gradually a trending research topic in both academics industry. Advancing application emerging field necessitates development new...