When: Monday, March 30, 2020
Where: Online presentation - Google Meet https://meet.google.com/imt-aboa-zne
Time: 11:00 AM
Focus on Research (FoR) is an opportunity for any COSC Ph.D. student to discuss a research project (with or without preliminary results), a conference dry run, or any research topic of interest to present to an audience of peers and faculty. It is a great avenue for Ph.D. students to practice presentation skills in front of a larger and broader audience.
Adversarial Deep Reinforcement Learning for Cyber-Physical System Security
Taha Eghtesad, Ph.D. Student
By compromising and tampering with the control of cyber-physical systems such as critical infrastructure, attackers could cause financial losses, physical damage, and even bodily harm. In addition to preventing cyber-attacks, defenders must also be prepared to promptly and effectively mitigate attacks that they could not prevent to minimize their impact. Eventually, defenders must secure compromised components by resetting them to a secure state (e.g., reinstalling and patching devices). However, securing compromised components is often not immediately possible (e.g., it may be impossible to reset compromised components remotely), which leaves attackers with enough time to cause damage. We focus on mitigating cyber-attacks against control systems immediately in their aftermath via adjusting the control policies of the system to compensate for adversarial tampering. In particular, we consider cyber-attacks that aim to cause damage by changing actuation signals and formulate the problem of finding resilient control policies that minimize the worst-case impact. We propose a learning-based computational approach for finding resilient control policies that can be deployed immediately to effectively mitigate a wide range of attack scenarios. We demonstrate the efficiency of our computational approach and the robustness of our control policies through simulations.Bio:
Taha Eghtesad is a second-year Ph.D. student of computer and information sciences at the University of Houston. He works under the supervision of Dr. Aron Laszka on cybersecurity and artificial intelligence. He holds a B.S. degree from Shahid Beheshti University, Tehran, Iran in computer engineering with a minor in software engineering.
Metamorphic Malware Detection Using Behavior Graphs
Ayman El Aassal, Ph.D. Student
Common malware detection techniques rely on malware signatures. However, these techniques are not able to catch new or mutated versions of malware. Hackers use different methods to mutate malware, and one of them is metamorphism. Metamorphic malware refers to a computer virus that changes its code after each iteration or before it propagates itself. Signature-based methods can’t detect this type of computer virus because their signatures keep changing with each iteration. The goal of my research is to trace the execution of such malware, graph its behavior, and use graph models to detect its mutations.
Ayman El Aassal is a third year Ph.D. student at the University of Houston. He received a masters degree in computer science in Morocco at the University of Computer Science (ENSIAS) in Rabat. His current field of research is in Cybersecurity, focusing on intrusion detection under the supervision of Dr. Stephen Huang.
Quasar: A Novel Density-Based Framework for Collocation Mining
Karima Elgarroussi, Ph.D. Student
With the emerging availability of geographical data, spatial data mining has become a popular field. In particular, for intelligent crime analysis, it helps understanding the dynamics of unlawful activities, providing possible answers to where, when and why certain crimes are likely to happen. For this reason, there is an urgent need for effective and efficient methods to extract collocation patterns from spatial datasets. Given a collection of boolean spatial features, the collocation pattern discovery process finds the subsets of features that are frequently located together. This project proposes a novel density-based collocation-mining framework called Quasar. It relies on collocation measures which estimate the strength of a giving collocated pattern. They are defined by combining non-parametric density functions, associated with a particular set of features in the longitude-latitude space. Global collocation patterns can then be identified by measuring their area under the curve, and regional collocation patterns can be obtained by identifying regions where the curve of the collocation measure is above a user-defined threshold. Furthermore, we generalize Quasar to mine collocation patterns in Spatio-temporal datasets using two approaches: The first approach generalizes collocation measures based on 3D density estimation; the second approach mines these Spatio-temporal collocation patterns using consecutive batches. Finally, we evaluate our approach in a real-world case study using various crimes of New York City.
Karima Elgarroussi is a third-year Ph.D. student at the University of Houston. She received a masters degree in e-management and business intelligence from the University of Computer Science (ENSIAS), Rabat, Morocco. Her current field of research is in data analysis and intelligent systems under the supervision of Dr. Christoph F. Eick.
Voxel Indexing and Compression
Mouad Rifai, Ph.D. Student
While the rapid growth in the availability and quality of airborne laser-scanning data offers unprecedented information, it challenges the existing data management solutions. Data management has become a bottleneck to effective laser-scanning data exploration. This talk describes some state-of-the-art techniques for mobile/aerial laser-scanning indexing and compression. The key strategies for data handling, including data modelling and indexing, are discussed.
Mouad Rifai is a Ph.D. student. His research revolves around data structures for voxel indexing and compression, and related energy-efficient specialized hardware designs and implementations. His doctoral advisor is Dr. Lennart Johnsson.