Recent studies demonstrated that the training process of deep neural networks (DNNs) is vulnerable to backdoor attacks if third-party resources (e.g., samples) are adopted. Specifically, adversaries intend embed hidden backdoors into DNNs, where can be activated by pre-defined trigger patterns and leading malicious model predictions. My dissertation focuses on poisoning-based in computer vision...

Research Purposes: The project is based on the traditional accounting audit platform, basis of technical research, to re-create a three-party relationship, widely used data platform. Methodology: We use SQL Server database storage principle, original documents and other first-hand information direct electronic storage. Results: It will not only increase file sharing privilege management algorit...

Abstract We use mathematical modeling methods to incorporate new elements into the computer graphics technology and visual communication design process. This method realizes of a image system. uses artificial vision model extracts local features. The micro-partial equation obtains background reference line target point in sequence complete suppression spatio-temporal domain. experimental result...