Generative graph self-supervised learning (SSL) aims to learn node representations by reconstructing the input data. However, most existing methods focus on unsupervised tasks only and very few work has shown its superiority over state-of-the-art contrastive (GCL) models, especially classification task. While a recent model been proposed bridge gap, performance is still unknown. In this paper, comprehensively enhance of generative SSL against other GCL models both supervised tasks, we...

Federated Learning (FL) has emerged as a de facto machine learning area and received rapid increasing research interests from the community. However, catastrophic forgetting caused by data heterogeneity partial participation poses distinctive challenges for FL, which are detrimental to performance. To tackle problems, we propose new FL approach (namely GradMA), takes inspiration continual simultaneously correct server-side worker-side update directions well take full advantage of server's...

Knowledge distillation on graph neural networks is a novel and practical model enhancement technique that has attracted widespread attention. However, literature methods mostly work in the manner of same teacher/student architecture. In fact, knowledge provided by identical teacher may be insufficient, leading to lack diversity consequently limiting capabilities student GNNs. this paper, we innovatively propose Distinct Multi-teacher Distillation method, namely DMKD, fully exploit...

In federated learning (FL), malicious clients could manipulate the predictions of trained model through backdoor attacks, posing a significant threat to security FL systems. Existing research primarily focuses on attacks and defenses within generic scenario, where all collaborate train single global model. A recent study conducted by Qin et al. [ 24 ] marks initial exploration personalized (pFL) each client constructs based its local data. Notably, demonstrates that pFL methods with...

A heterogeneous information network (HIN) is one whose nodes model objects of different types and links objects' relationships. To enrich its information, in an HIN are typically associated with additional attributes. We call such <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Attributed HIN</i> or AHIN. study the problem clustering AHIN, taking into account similarities respect to both object attribute values their structural connectedness...

In recent years, self-supervised learning has emerged as a promising approach in addressing the issues of label dependency and poor generalization performance traditional GNNs. However, existing methods have limited effectiveness on heterophilic graphs, due to homophily assumption that results similar node representations for connected nodes. this work, we propose multi-view contrastive model namely, MUSE. Specifically, construct two views capture information ego its neighborhood by GNNs...

We propose knowledge internalization (KI), which aims to complement the lexical into neural dialog models. Instead of further conditioning knowledge-grounded (KGD) models on externally retrieved knowledge, we seek integrate about each input token internally model's parameters. To tackle challenge due large scale adopt contrastive learning approach and create an effective token-level retriever that requires only weak supervision mined from Wikipedia. demonstrate effectiveness general...

We study the problem of applying spectral clustering to cluster multi-scale data, which is data whose clusters are various sizes and densities. Traditional techniques discover by processing a similarity matrix that reflects proximity objects. For distance-based not effective because objects sparse could be far apart while those dense have sufficiently close. Following [16], we solve on integrating concept objects' "reachability similarity" with given derive an coefficient matrix. propose...

Monocular Semantic Occupancy Prediction aims to infer the complete 3D geometry and semantic information of scenes from only 2D images. It has garnered significant attention, particularly due its potential enhance perception autonomous vehicles. However, existing methods rely on a complex cascaded framework with relatively limited restore scenes, including dependency supervision solely whole network's output, single-frame input, utilization small backbone. These challenges, in turn, hinder...

Federated learning (FL) on heterogeneous data (non-IID data) has recently received great attention. Most existing methods focus studying the convergence guarantees for global objective. While these can guarantee decrease of objective in each communication round, they fail to ensure risk client. In this paper, we propose FedCOME, which introduces a consensus mechanism aiming decreased client after training round. particular, allow slight adjustment client's gradient server-side, producing an...

The task of vision-based 3D occupancy prediction aims to reconstruct geometry and estimate its semantic classes from 2D color images, where the 2D-to-3D view transformation is an indispensable step. Most previous methods conduct forward projection, such as BEVPooling VoxelPooling, both which map image features into grids. However, current grid representing within a certain height range usually introduces many confusing that belong other ranges. To address this challenge, we present Deep...

Text-to-video generation has evolved rapidly in recent years, delivering remarkable results. Training typically relies on video-caption paired data, which plays a crucial role enhancing performance. However, current video captions often suffer from insufficient details, hallucinations and imprecise motion depiction, affecting the fidelity consistency of generated videos. In this work, we propose novel instance-aware structured caption framework, termed InstanceCap, to achieve instance-level...

A heterogeneous information network (HIN) has as vertices objects of different types and edges the relations between objects, which are also various types. We study problem classifying in HINs. Most existing methods perform poorly when given scarce labeled training sets, that improve classification accuracy under such scenarios often computationally expensive. To address these problems, we propose ConCH, a graph neural model. ConCH formulates multi-task learning combines semi-supervised with...

The initial clustering center of traditional K-means algorithm is selected at random that different will get results, which have great randomicity and poor stability.To improve the optimized by adopting local outlier index, we adopt a positive approach calculating index all data samples.Then k dense points with furthest mutual distance were as center.At last, in this paper eliminate effects using improved algorithm.The experimental results showed enhanced could reduce susceptibility to...

Session-based recommendation (SBR) is a task that aims to predict items based on anonymous sequences of user behaviors in session. While there are methods leverage rich context information sessions for SBR, most them have the following limitations: 1) they fail distinguish item-item edge types when constructing global graph exploiting cross-session contexts; 2) learn fixed embedding vector each item, which lacks flexibility reflect variation interests across sessions; 3) generally use...

While generating better negative samples for contrastive learning has been widely studied in the areas of CV and NLP, very few work focused on graph-structured data. Recently, Mixup introduced to synthesize hard graph (GCL). However, due unsupervised nature GCL, without help soft labels, directly mixing representations could inadvertently lead information loss original further adversely affect quality newly generated harder negative. To address problem, this paper, we propose a novel method...