基于深度情景群智感知的面向移动互联网的主动防御技术研究

Mobile security problem is becoming more and more serious. In order to effectively perceive the mobile context for active defense, Crowd Sensing provide large data and low cost labels. However, traditional learning methods failed to adapt to its dynamic and mixed data environment. To solve above problem, we study the Key theories and methods of Deep Crowd-sensing based Mobile Context-Aware Active DefenseTechnique, which including Multi-source multimodal sensor data association and fusion based on Deep Mutual transfer learning; Constrained online incremental learning based scene discovery method in Mobile Crowd-Sensing; Dynamic adaptive context aware access control model based on automatic classification combined with user feedback. This project aims to reveal the "cross source and cross modal Deep Mutual transfer" optimal rules between the context aware model in the crowd sensing, realization of accurately context classification and deep fusion of the mixed context data; and to find an effective deep incremental scene discover method and context security grading method, and final realization of mobile security context aware dynamic access control policy. It is expected to be a breakthrough in the context aware active security technology in the mobile environment, will also has theoretical significance of considerable support to Crowd-sensing problems of common concern.

移动安全问题日益严重，准确有效地感知所处情景态势是实施主动防御策略的前提。移动互联网的群智感知为改进情景分类性能提供大规模数据和低成本标记，然而面向主动防御的传统情景感知方法难以应对其动态、混杂、非安全大数据环境。本项目针以上问题，研究基于深度群智感知的面向移动互联网的主动防御关键理论和方法，包括基于互迁移学习的群智感知数据深度融合；深度群智动态感知中有约束的情景发现与在线增量自学习；自动情景分类结合用户反馈的动态自适应访问控制模型；情景数据的隐私保护哈希深度学习。本项目旨在揭示群智条件下混杂情景数据的“深度互迁移”优化融合规律并实现高效分类；并找到有效的深度情景发现增量机制和情景安全属性划分方法，以及安全有效的群智感知隐私保护手段；最终实现群智环境下动态精准的情景感知主动防御策略。预期将在移动安全态势大数据感知等主动防御关键技术上形成突破，也将对群智感知中的共性问题具有重要价值。

移动安全问题日益严重，准确有效地感知所处情景态势是实施主动防御策略的前提。基于此，本项目提出了多种基于深度群智感知的面向移动互联网的主动防御的关键理论和方法，在本项目资助下，我们在网络与信息安全领域、相关智能计算领域上发表了SCI检索论文40余篇，EI检索论文50余篇，包括IEEE Transactions on Information Forensics and Security、IEEE Internet of Things Journal、IEEE Transactions on Pattern Analysis and Machine Intelligence、IEEE Transactions on Cybernetics等；授权或申请国家发明专利10余项；培养博士、硕士研究生20余名；参加了多个学术会议，并进行了广泛的学术交流。

内容获取失败，请点击重试