Research Projects

Abstract

The fundamental goal of money laundering is to make criminal income appear as if it has been derived from a legitimate source. Cybercriminals use the sharing economy platforms, including ridesharing, short-term rentals, the gig economy, on-demand delivery, peer-to-peer lending, crowdfunding, reselling and trading, and even entertainment and video games, to launder their ill-gotten gains. The AML research group draws on previous work conducted in criminology and is working toward novel ways to detect criminal activity.

Professors

• Stephen Flowerday, Ph.D.
• Andrew Morin, Ph.D.
• Weiping Pei, Ph.D.

Abstract

To explore the challenges organizations face when running their own SOC or using a virtual SOC provided by a managed security service provider. What makes this research different is the focus on the interplay between organizational, human, and technical challenges present in operating a successful SOC.

Professors

• Tyler Moore, Ph.D.
• Stephen Flowerday, Ph.D.

Abstract

The goal of this project is to detect anomalies on networks with time-varying node or edge properties. Applications include detection of criminal transactions on cryptocurrency networks, rewiring of biological networks (e.g., neural and gene networks) in disease states, earthquake detection from seismograph sensor networks, and detection of science events on autonomous networks of cooperative orbiters. Methodologies include network machine learning, kinetic (differential equation) models on networks, extreme value theory and Kalman filters.

Professors

• Brett McKinney, Ph.D. 

Abstract

This project aims to create a viable framework for delivering different types of VR training simulations to educators to develop simulations that have multiple correct answers, ordered and unordered series of steps, and valued accuracy of each procedure.

Professors

• Akram Taghavi-Burris
• John Hale, Ph.D.

Abstract

This project will focus on developing attack graphs showing how a system can be compromised to build and deploy cyber defense tools that continuously monitor the system and adapt to changing conditions.

Mentor

• Peter Hawrylak, Ph.D.

Abstract

Building on prior research on trusted human-AI collaboration and transparent decision-making, this project develops two complementary frameworks in the increasingly critical area of Explainable Artificial Intelligence (XAI): Personalized Explanation Systems (Perx) and Explaining Options & Recommendations (EXPLORE). Medical device application (defect detection and damage quantification in Nitinol) provides the unique data source needed to prototype the framework.

Professors

• Sandip Sen, Ph.D.
• Will LePage, Ph.D.

Abstract

This project proposes a novel, interpretable, and trustworthy machine learning framework that detects fault and cyber attack incidents associated with the electric power grid and its recovery from critical system incidents in real-time.

Mentor

• Mahdi Khodyar, Ph.D.

Abstract

Robots and autonomous systems most often use servomotors to generate the motion required to do their tasks. The control is simple, but a different servomotor (or other actuator) needs to be sized and selected for each and every degree of freedom, which consumes a lot of person-hours of effort. If a back-up system is needed, it is heavy and bulky, and the entire process may need to be repeated again. This cyber-fellows project investigates a different paradigm inspired by human muscles: each degree of freedom is actuated by a collection of modular units that work together the way muscle cells do, with each module being on-off only. In this way only a single part number needs to be tracked and stocked, and adjustments and repairs can be made on the fly. Redundancy can be built-in by including extra modules in the design. However, controlling a system like this is complicated because activation of each module needs to be coordinated with all the others. This project investigates ways to predict how collections of modules will behave in order to plan their structure and activations accordingly.

Professors

• Joshua A. Schultz, Ph.D.

Abstract

The project designs and implements an attack-resilient decentralized multiagent framework to enhance the reliability and resilience of smart grids and large-scale power networks. The presented work estimates the type and severity of the attacks using cutting-edge graph learning architectures and deep neural networks. Moreover, the project incorporates physics rules of the electric network into deep hierarchical reinforcement learning to provide an automatic approach for real-time power system recovery and control. The knowledge interpretation and human interaction modules would help the power system operators to visualize the control/recovery policies, validate them, and update them based on their expertise.

Professors

• Mahdi Khodayar, Ph.D.

Abstract

The trend to connect and link supervisory control and data acquisition (SCADA) systems to the Internet provides numerous benefits, but at the same time it leaves them open to cyber attacks. SCADA systems oversee most of the critical infrastructure (e.g., power grid and water utilities) and their interruption is a serious problem for society. The vast majority of existing SCADA cyber security tools are essentially IT security tools that have been retrofitted to protect a process control network. Furthermore, these tools are often configured and maintained by IT personnel who may have limited knowledge of the control process itself. There is a need for tools that effectively apply accepted network security practices to SCADA systems but also take into account process safety requirements as specified by a process engineer. Such tools would be capable of detecting and reacting to new forms of attacks, such as Stuxnet, that take advantage of the physical process and the disconnect between the process engineers and IT personnel to compromise the overall system. A new framework is proposed that takes input from both IT security professionals and process engineers to produce actionable output, i.e., output that can be immediately used by existing network security components such as firewalls, IDS (e.g., Zeek, Suricata, Snort) and others.

Professors

• Mauricio Papa, Ph.D.
• Peter Hawrylak, Ph.D.

Abstract

This project seeks to explore and resolve ethical, regulatory and legal obstacles to responsible use of AI and ML in both the government and business intelligence communities’ rush to harvest and exploit data for intelligence. It will expose the need for these stakeholders to respect these considerations in using these tools, and offer a holistic approach for managing and controlling their use. The Intelligence Community’s (IC’s) analytical software tools and processes are designed to support the Intelligence Cycle, a process with strict but dated compartmentalized standards. This community’s primary concern is national security, while the BI community focuses on organizational performance. However, both can learn from each other concerning their ethical and operational standards for engaging intelligence analysis tools. These tools apply AI and ML techniques, capturing and synthesizing multiple data streams from largely open-source origins. Yet the new ‘data lakes’ created by the synthesis of such data sources are not under the same governance as the original data. Further, every corresponding process and application must meet the data ownership or custodial duty obligations of the new data lake.

Professors

• John Hale, Ph.D.

Abstract

The Theory of Inventive Problem Solving (TRIZ) method describes a process to focus innovation and research efforts to achieve significant, “game changing”, advances in a given field. The method focuses on breaking down problems and questions to their most simple innovative and technical questions. At this point the method investigates contradictions within the design space and ranks these to help innovators identify the best paths to pursue to achieve “game changing” advances verses “incremental” advances. Much of this process is still manual and there are limited databases available to provide input for applying the method and analysis of existing IP (e.g., patents and patent applications) for further study. Automating this process would be of significant value to a company and the larger society. This project will focus on two goals: (1) developing an automatic TRIZ classification system to classify patents according to the 40 Inventive TRIZ Principles and the 39 Engineering TRIZ Parameters used to generate the TRIZ contradiction matrix; and (2) developing a natural language processing ingestion tool to convert the patents into the format needed for the classification system.

Professors

• Peter Hawrylak, Ph.D.
• John Hale, Ph.D.

Abstract

Building on our prior research on trusted human-AI collaboration, surrogate and supportive agents, learning and adaptation in agent networks, user modeling, and explainable AI, the PGAs project is a framework for designing, developing, and implementing personalized AI/ML assistants that can guide users to make effective decisions when selecting between alternative plan of actions to achieve their goals. We note that plan alternative can significantly vary in efficacy, costs, associated risks, and success rates. Plan selection may often require careful tradeoffs. Human decision makers have knowledge gaps, ingrained biases, established preferences, as well as tolerance and aversion for uncertainty and risks. Combination of and interaction between these myriad factors can leave individual decision makers, both in their personal and professional environments, susceptible to making sub-optimal decisions that affect performance and satisfaction as well as leaving them vulnerable to various threats from malevolent actors undermining their safety and security. PGAs are designed to be knowledgeable support staff to human decision makers who provide both preferred action plans to achieve goals and mitigate risks and threats as well as the associated rationale to persuade the user to adopt them.

Professors

• Sandip Sen, Ph.D.

Abstract

The proposed effort will yield an architecture and reference implementation of a cyber range for education and training programs focused on cybersecurity, but extensible to other cyber domains. The cyber range will incorporate a platform for content delivery to students, and also deeply embed instrumentation to measure the effectiveness of curricular elements. An automation engine will engage instrumentation components and link them to an assessment module that maps gathered measures to assessment criteria articulated by educators.

Professors

• John Hale, Ph.D.