• Neural Network Backdoor Threat in Industrial Control Systems [RICCS in CCS'24] [pdf]

The neural network data-free backdoor threat poses a significant risk to industrial control systems (ICS), where increasing integration of neural network applications within systems heightens their vulnerabilities. This study represents preliminary work in understanding the data-free backdoor attack and assessing its potential impact on ICSs. The key factors influencing attack performance are examined through experimental research. Potential risks specific to ICS are identified through threat modeling. The insights gained could inform the development of more robust defense strategies, enhancing the protection of critical ICS infrastructure against these neural network backdoor threats.

Other work related to Cloud Native Services and Manufacturing: 

• The Development of A Large-Scale Cloud Emulator [IC2E'24] [pdf]

• Security Challenges for Modern Data Centers with IoT: A Preliminary Study [WebConf'22] [pdf]

• Enhancing Order Fulfillment Through Production Process Reengineering Using Manufacturing Execution System as a Reference Model [IEEE Access,2024] [pdf]

• Facilitating Threat Modeling by Leveraging Large Language Models [AISCC in NDSS'24] [pdf]

This preliminary work developed a novel GenAI-Assisted threat modeling system by leveraging the Llama and RAG pipeline to decrease the required human effort in the threat modeling process. Two major threat modeling questions are considered in the proposed task workflows, where the NLP techniques assist in parsing and understanding design documents and threats, and the LLM analyzes and synthesizes volumes of documentation to generate responses to related threat modeling questions. 

• ILLATION: Learning Vulnerability Risk from Network [IEEE TDSC, 2023] [pdf]

Enable AI-powered network-specific vulnerability risk prioritization to support efficient vulnerability patching by utilizing a neural network and logic reasoning AI model to learn and infer adversaries' motivation and ability in a network while learning the constraints that restrict interactions between vulnerabilities and network elements.

• LICALITY: Learning Vulnerability RIsk From Attacker [IEEE TNSM, 2021] [pdf]

Addressing the limitation of the Common Vulnerability Scoring System (CVSS) on network-agnostic vulnerability risk prioritization by capturing the attacker's preference for exploiting vulnerabilities through the proposed threat modeling method, and learning the associated threat attributes by utilizing neuro-symbolic AI technique in a developed neural network - probabilistic logic programming (NN-PLP) model.

• AI-Powered Cybersecurity Cloud Hands-on Lab Platform [IEEE FIE'18, ACM SigCSE'18, ITiCSE'21, JAIT'21] 

The developed virtual hands-on lab platform is designed for computer science education to support personalized learning by utilizing the techniques of cloud computing. Student learning styles can be identified from student activities. With the awareness of student learning styles, instructors are able to use techniques more suitable for students, and hence, improve student overall learning experiences. By developing the lab context-based knowledge graph, a personalized learning plan is provided to support student learning on the online virtual hands-on lab platform.  It utilizes natural language processing (NLP) techniques to construct the knowledge graph from lab contents associated with cybersecurity topics, which guides learners to work on cybersecurity lab projects independently.  

* This research is supported by the NSF EAGER: SaTC-EDU: Artificial Intelligence for Cybersecurity Education via a Machine Learning-Enabled Security Knowledge Graph and NSF SaTC: EDU: Learning Moving Target Defense Concepts: Teaching and Training Curricula Development Based on Software Defined Networking and Network Function Virtualization ( Zeng worked on this project as Ph.D. student)