SaTC: CORE: Small: Collaborative: A Framework for Enhancing the Resilience of Cyber Attack Classification and Clustering Mechanisms

分类和聚类是两类重要的机器学习技术，已被广泛用于网络防御目的。然而，这些机制可以被智能规避攻击击败，例如对抗性机器学习。目前，还没有有效的对策来对付这些复杂的攻击。该项目的目标是研究有效的对策，使分类和集群机制能够抵御智能规避攻击。该项目的科学贡献包括促进我们对规避攻击的可行性和影响的理解，以及设计对此类攻击具有鲁棒性的机器学习算法。由于机器学习技术广泛应用于许多其他领域，如现实世界的欺诈和犯罪检测，这些领域也将受益于该项目。该项目将涉及博士生，他们将直接为下一代劳动力做出贡献，并将通过让女学生和代表性不足的学生参与进来来解决多样性问题。该项目计划通过调查一种比文献中调查的传统黑箱攻击更强大的攻击类别（称为灰箱攻击）来实现其目标。在灰盒攻击模型中，攻击者可以执行防御者通常执行的所有活动。该项目将建立一个理论模型和框架，用于表征灰箱模型下分类和聚类机制对智能规避攻击的脆弱性和弹性，增强分类和集群机制，以抵御智能规避攻击，并获得可量化的弹性收益。该奖项反映了NSF的法定使命，并通过使用基金会的知识产权进行评估，被认为值得支持。优点和更广泛的影响审查标准。

Classification and clustering are two important classes of machine learning techniques that have been widely used for cyber defense purposes. However, these mechanisms can be defeated by intelligent evasion attacks, such as Adversarial Machine Learning. Currently, there are no effective countermeasures against these sophisticated attacks. The objective of the project is to investigate effective countermeasures to make classification and clustering mechanisms robust against intelligent evasion attacks. The scientific contributions of the project include advancing our understanding of the feasibility and impact of evasion attacks, and the design of machine learning algorithms that are robust against such attacks. Since machine learning techniques are widely employed in many other areas such as real-world fraud and crime detection, those areas would benefit from this project too. The project will involve PhD students who will directly contribute to the next-generation workforce and will address diversity by involving female students and students from underrepresented groups.The project plans to achieve its goal by investigating a more powerful class of attacks, called gray-box attacks, than the traditional black-box attacks investigated in the literature. In the gray-box attack model, the attacker can perform all the activities that a defender would normally perform. The project will build a theoretical model and framework for characterizing the vulnerability and resilience of classification and clustering mechanisms with respect to intelligent evasion attacks under the gray-box model, enhance classification and clustering mechanisms to withstand intelligent evasion attacks with quantifiable resilience gains.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.