Graph neural network (GNN) has captured wide attention due to its capability of graph representation learning for graph-structured data. However, the distributed data silos limit the performance of GNN. Vertical federated learning (VFL), an emerging technique to process distributed data, successfully makes GNN possible to handle the distributed graph-structured data. Despite the prosperous development of vertical federated graph learning (VFGL), the robustness of VFGL against the adversarial attack has not been explored yet. Although numerous adversarial attacks against centralized GNNs are proposed, their attack performance is challenged in the VFGL scenario. To the best of our knowledge, this is the first work to explore the adversarial attack against VFGL. A query-efficient hybrid adversarial attack framework is proposed to significantly improve the centralized adversarial attacks against VFGL, denoted as NA 2 , short for Neuron-based Adversarial Attack. Specifically, a malicious client manipulates its local training data to improve its contribution in a stealthy fashion. Then a shadow model is established based on the manipulated data to simulate the behavior of the server model in VFGL. As a result, the shadow model can improve the attack success rate of various centralized attacks with a few queries. Extensive experiments on five real-world benchmarks demonstrate that NA 2 improves the performance of the centralized adversarial attacks against VFGL, achieving state-of-the-art performance even under potential adaptive defense where the defender knows the attack method. Additionally, we provide interpretable experiments of the effectiveness of NA 2 via sensitive neurons identification and visualization of t-SNE. The code and datasets are available at https://github.com/hgh0545/NA2.

No abstract

No abstract

-Online school is complete by Internet off-campus teaching process; As long as have Internet conditions can at any time, anywhere access to the Internet, to choose their own learning. Relative to the face-to-face education, network education maximum breakthrough the limit of time and space education, is the separation between teachers and students, the teaching activities organized by the face to face, is a kind of across schools, cross-regional education system and teaching mode. While the hj operations is one of the winners.Index Terms -online school, network education, hj operations BACKGROUNDSeparation of teacher and students; With specific transmission system and media teaching; There are various ways of information transmission; Learning form of space and flexible. It can break through the limit of space and time; Provide more learning opportunities; Expand the scale of teaching; To improve teaching quality. Reduce the cost of teaching. Based on the characteristics and advantages of network education, online school has become a mainstream trend of learning.A. Construction Purpose 1) Establish the school's online business card, and school education based on Internet brand image; Establish the school's online business card, and school education based on Internet brand image;2) To establish a convenient online sharing of teaching resources and information platform, to realize the effective integration of education teaching resources;3) Set up smooth communication between students and teachers, parents and schools, interaction's interactive platform; 4) Good information display form, the realization of general information, information transparency; 5) Realize the student information (grades, one's status as a student), teaching equipment, teaching resource network management; 6) To establish a variety of teaching bases, the realization of a variety of means of auxiliary teaching (such as: video teaching, PPT); 7) Many kinds of ways of teachers, students, schools, departments of research papers, scientific research achievements; 8) Deepen to strengthen the interaction between parents and schools, promote teaching quality in common; 9) Perfect school net school security mechanisms to ensure information security and readability; 10) Improve the user to browse and use experience; 11) Improve the quality of lighting information and content. B. Advantage in Resource utilizationVarious education resources through the network spans the limitation of space distance, make the school education can go beyond the campus radiation to a broader region of open education. Schools can give full play to their discipline and education resource advantages, the most outstanding teachers, the best teaching results through the network to spread to all directions. C. Learning behaviorNetwork technology application of remote education, its characteristic is: anyone, any time, any place, from any chapter, studying any course. Network education is convenient and flexible "five no", in the learning mode in the most directly refle...

CAPTCHA is a completely automated program designed to distinguish whether the user is a computer or human. As the problems of Internet security are worsening, it is of great significance to do research on CAPTCHA. This article starts from the recognition of CAPTCHAs, then analyses the weaknesses in its design and gives corresponding recognition proposals according to various weaknesses, finally offers suggestions related to the improvement of CAPTCHAs. Firstly, this article briefly introduces the basic steps during the decoding process and their principles. And during each step we choose methods which are better adapted to the features of different CAPTCHA images. Methods chosen are as followings: bimodal method in binarization, improved corrosion algorithm in denoising, projection segmentation method in denoised image processing and SVM in recognition. Then, we demonstrate detailed process through the samples taken from the online registration system of ICBC, show the recognition effect and correct the results according to the statistical data in the process. This article decodes CAPTCHAS from three other large banks in the same way but just provides the recognition results. Finally, this article offers targeted suggestions to the four banks based on the recognition effect and analysis process stated above.