Machine learning enabled adaptive user interfaces for enhanced network security

在日益复杂和新颖的网络攻击形式下，人工网络监控对于网络安全至关重要。不幸的是，当所有信息同时可视化时，跨计算机网络的不同逻辑视图手动监控设备的复杂配置会导致信息过载。为此，关键是构建自适应网络可视化工具，以最少的交互来预测用户的需求，并在不使用户过载的情况下对潜在的网络威胁进行分类。这种适应性的核心是两个任务：(1)需要在系统状态和相关风险的动态影响图模型中执行快速推理，以告知界面适应性；(2)需要学习并持续更新系统模型和用户模型，以确保在第(1)部分中进行准确的推理和决策。我们将与Uncharted合作开发新的机器学习和推理算法，以在一个新的界面中同时处理任务(1)和(2)，以开发新的增强型网络安全监控工具。虽然这项研究项目将直接应用于网络安全和网络监测，但基本技术应该推广到广泛的自适应用户界面。

Manual network monitoring is essential for network-security in the presence of increasingly sophisticated and novel forms of cyber-attack. Unfortunately, manually monitoring complex configurations of devices across different logical views of a computer network leads to an information overload when all information is visualized simultaneously. To this end, it is critical to build adaptive network visualization tools that anticipate users' needs with minimal interaction as well as triage potential network threats without overloading a user. At the core of such adaptivity are two tasks: (1) the need to perform fast inference in dynamical influence diagram models of system state and associated risks to inform interface adaptations and (2) the need to learn and continually update both the system model and the user model in order to ensure accurate inference and decision-making in part (1). We will partner with Uncharted to develop novel machine learning and inference algorithms to address both tasks (1) and (2) in a novel interface for the development of novel enhanced network security monitoring tools. While this research project will have direct application to cyber-security and network monitoring, the underlying techniques should generalize to a broad range of adaptive user interfaces.