Graph data widely exist in many high-impact applications. Inspired by the success of deep learning grid-structured data, graph neural network models have been proposed to learn powerful node-level or graph-level representation. However, most existing networks suffer from following limitations: (1) there is limited analysis regarding convolution properties, such as seed-oriented, degree-aware and order-free; (2) node's degreespecific structure not explicitly expressed for distinguishing...

Salp Swarm Algorithm (SSA) is a novel swarm intelligent algorithm with good performance. However, like other swarm-based algorithms, it has insufficiencies of low convergence precision and slow speed when dealing high-dimensional complex optimisation problems. In response to this concerning issue, in paper, we propose an improved SSA named as WASSA. First all, dynamic weight factor added the update formula population position, aiming balance global exploration local exploitation. addition,...

Recommender systems are powerful tools for information filtering with the ever-growing amount of online data. Despite its success and wide adoption in various web applications personalized products, many existing recommender still suffer from multiple drawbacks such as large unobserved feedback, poor model convergence, etc. These work mainly due to following two reasons: first, widely used negative sampling strategy, which treats unlabeled entries samples, is invalid real-world settings;...

Federated learning (FL) naturally faces the problem of data heterogeneity in real-world scenarios, but this is often overlooked by studies on FL security and privacy. On one hand, effectiveness backdoor attacks may drop significantly under non-IID scenarios. other malicious clients steal private through privacy inference attacks. Therefore, it necessary to have a comprehensive perspective heterogeneity, backdoor, inference. In paper, we propose novel inference-empowered stealthy attack...

With decentralized data collected from diverse clients, a personalized federated learning paradigm has been proposed for training machine models without exchanging raw local clients. We dive into the perspective of privacy-preserving transfer learning, and identify limitations previous algorithms. First, works suffer negative knowledge transferability some when focusing more on overall performance all Second, high communication costs are required to explicitly learn statistical task...

Unsupervised domain adaptation has been successfully applied across multiple high-impact applications, since it improves the generalization performance of a learning algorithm when source and target domains are related. However, adversarial vulnerability models largely neglected. Most existing unsupervised algorithms might be easily fooled by an adversary, resulting in deteriorated prediction on domain, transferring knowledge from maliciously manipulated domain.

Federated learning learns a neural network model by aggregating the knowledge from group of distributed clients under privacy-preserving constraint. In this work, we show that paradigm might inherit adversarial vulnerability centralized network, i.e., it has deteriorated performance on examples when is deployed. This even more alarming federated designed to approximate updating behavior network. To solve problem, propose an adversarially robust framework, named Fed_BVA, with improved server...

Imbalanced data widely exist in many high-impact applications. An example is air traffic control, where among all three types of accident causes, historical reports with `personnel issues' are much more than the other two (`aircraft and `environmental issues') combined. Thus, resulting set highly imbalanced. On hand, this can be naturally modeled as a network, each node representing an report, edge indicating similarity pair reports. Up until now, most existing work on imbalanced analysis...

Abstract Background and Objectives In the process of microwave drying rice, dielectric property determines ability rice to absorb microwave, change constant loss factor has a great influence on rice. Therefore, it is necessary predict Findings This study investigated intermediate electrical characteristics during drying. It explored their relationship with temperature, water content, frequency. A mathematical model was developed at frequencies 2.45 GHz 500–3000 MHz constant, factor....

Open-set domain adaptation aims to improve the generalization performance of a learning algorithm on target task interest by leveraging label information from relevant source with only subset classes. However, most existing works are designed for static setting, and can be hardly extended dynamic setting commonly seen in many real-world applications. In this paper, we focus more realistic open-set time evolving where novel unknown classes appear over time. Specifically, show that...

In order to solve the problems of poor retrieval performance and low quality in traditional method, a non-metallic pipe system based on ontology was designed. Firstly, theory information technology were introduced, then an model that can describe different oilfield built. The files type html described by OWL language. After that, they saved My SQL database Jena. Secondly introduced algorithm process system, realized semantic analysis query condition, expanded other functions. At last, domain...

Networks are ubiquitous in many real-world applications due to their capability of representing the rich information data. One fundamental problem network analysis is learn a low- dimensional vector representation for nodes within attributed networks. However, there little work theoretically considering heterogeneity from networks, and most existing embedding techniques able capture at k-th order node proximity, thus leading loss long-range spatial dependencies between individual across...

In this paper, we present a statistical analysis of six traffic features based on entropy and distinct feature number at the packet level, find that, although these are unstable show seasonal patterns like volume in long-time period, they stable consistent with Gaussian distribution short-time period. However, equilibrium property will be violated by some anomalies. Based observation, propose Multi-dimensional Box plot method for Short-time scale Traffic (MBST) to classify abnormal normal...

Transfer learning has been successfully applied across many high-impact applications. However, most existing work focuses on the static transfer setting, and very little is devoted to modeling time evolving target domain, such as online reviews for movies. To bridge this gap, in paper, we study a novel continuous setting with domain. One major challenge associated potential occurrence of negative domain evolves over time. address challenge, propose label-informed C-divergence between source...

Graph data widely exist in many high-impact applications. Inspired by the success of deep learning grid-structured data, graph neural network models have been proposed to learn powerful node-level or graph-level representation. However, most existing networks suffer from following limitations: (1) there is limited analysis regarding convolution properties, such as seed-oriented, degree-aware and order-free; (2) node's degree-specific structure not explicitly expressed for distinguishing...

Transfer learning refers to the transfer of knowledge or information from a relevant source task target task. However, most existing works assume both tasks are sampled stationary distribution, thereby leading sub-optimal performance for dynamic drawn non-stationary distribution in real scenarios. To bridge this gap, paper, we study more realistic and challenging setting with tasks, i.e., continuously evolving over time. We theoretically show that expected error on can be tightly bounded...

Heterogeneous data widely exists in various high-impact applications. Domain adaptation and out-of-distribution generalization paradigms have been formulated to handle the heterogeneity across domains. However, most existing domain algorithms do not explicitly explain how label information can be adaptively propagated from source domains target domain. Furthermore, little effort has devoted theoretically understanding convergence of based on neural networks.To address these problems, this...

Networks are natural representations in modeling adversarial activities, such as smuggling, human trafficking, and illegal arms dealing. However, activities often covert embedded across multiple domains sources. They generally not detectable recognizable from the perspective of an isolated network, only become apparent when networks analyzed a unified m anner. T o t his e nd, we propose Complex Analytics Network (CANON), mathematical computational framework for large-scale, multi-sourced...

In federated learning (FL), multiple clients collaborate to train machine models together while keeping their data decentralized. Through utilizing more training data, FL suffers from the potential negative transfer problem: global model may even perform worse than trained with local only. this paper, we propose FedCollab, a novel framework that alleviates by clustering into non-overlapping coalitions based on distribution distances and quantities. As result, each client only collaborates...