SFS Research

Faculty-Guided Research

Research absracts from faculty mentor and SFS scholars. 

Advisor: Dr. Jie Yan

Dr. Jie Yan is a professor in the Department of Computer Science and the Director of Machine Perception & Animation lab and Cybersecurity Application Lab at Bowie State University. She is also the faculty lead for center for Cyber Security and Emerging Technologies.  Her areas of expertise include Computer Graphics, Computer Vision, Machine Learning, and Cybersecurity Education.

Scholars: Jada A Danner, Kamal A EPPS

Title: Differentiating Real of Authentic Images from Fake Imagery

Research Abstract:

Differentiating real or authentic images from manipulated or fake imagery is significant for image forensics domain and the public. Researchers tend to develop very complicated deep learning architectures to train detector models making the process of fake image detection expensive to the public and limiting model usability.  

The goals of  this research project are : 1) to develop a robust forensic analyzer to detect authentic from counterfeit images using deep learning and a less expensive lightweight device with less computing power, 2) to produce a Raspberry Pi-based analyzer model that will be accessible to the general public,  3) to determine the best framework for image classification using raspberry pi, 4) to evaluate the performance of three frameworks in terms of image classification accuracy and model inference speed on IoT raspberry Pi device and, 5) to evaluate the performance of three types of strategies in terms of average detection accuracy and model complexity based on the best-performed framework. 

Advisor: Dr. Sarker Kamruzzman

Scholars: Roxan C. Rockefeller, Dawn L. Marshall

Title: Malware detection using Deep Learning 

Abstract:

Task: Use deep learning approach to detect malware.  

Specifics: Malware detection is an important part of the cybersecurity/secure computing. There are two main categories for classifying malware detection approaches: static analysis and dynamic analysis. Each of these categories has its own strengths and weaknesses. For instance, static analysis is quicker, but it cannot identify malware variants produced through code obfuscation. On the other hand, dynamic analysis is capable of detecting such variants effectively, but it operates at a slower pace and demands substantial resources. Existing solutions can use static, dynamic or a hybrid approach.  

In this project we will try to investigate existing solutions to detect malware and if time permits, we will use transformer based deep learning approach to explore how accurately it can detect malware. We will use existing malware dataset to test our algorithm.  

 Sample dataset:  

• https://www.kaggle.com/c/malware 

classification/data  

• https://virusshare.com/  

Goal:  

By the end of the 14 weeks (the fall semester, 2023), our exploration will shade lights on these: 

• Get idea about the existing solutions of malware detection. 
• The effectiveness of transformer based deep learning approach for malware detection over the other methods.  

Learning outcome: 

• Understand the difficulties of malware detection.  
• Explore existing solutions for the malware detection. 
• Employ transformer based deep learning algorithms for malware detection.  
• Gain hands on experience on cybersecurity. 

Advisor: Anijoy Chakma

Anijoy Chakma: is an incoming Assistant Professor in the Computer Science department at Bowie State University. He completed his Ph.D. in Information Systems from the University of Maryland, Baltimore County, in July 2023. He received a Bachelor of Science degree in Computer Science and Engineering from the Bangladesh University of Engineering and Technology in 2013 and a Master of Science degree in Computer Science from Lamar University, Texas, in 2018. Before his graduate study, he had 2.5 years of experience working in the software industry. He has published multiple peer-reviewed journals and conference papers, and he is the recipient of the Best Paper Award at the IEEE/ACM CHASE Conference 2022. During his Ph.D., he actively supervised multiple Summer Research Experience for Undergraduates (REU) students. His research interests are cyber-physical systems, smart health, and machine learning. 

Scholar: Jared L. Robinson

 

Title: Cross-domain Sensing & Modeling for Smart Environments 

Abtract:

Cyber-physical systems that create smart environments are associated with multiple sensing 

mechanisms to interact with and understand the physical world. Efficiently aggregating and processing multi-stream data and developing adaptive machine learning models to make actionable decisions is not trivial. Often, smart sensing (such as smartphones, smartwatches, Google Home, etc.) device provided data sources differ significantly from one data source to another.  

Most existing machine learning approaches need to be more adaptive and scalable to handle such variations. Therefore, developing scalable machine learning models under the presence of multi-stream, multi-modal data sources with labeled and unlabeled information is a challenging avenue to investigate. In this talk, I will discuss the underlying causes of data source variations, how to overcome such data variations to process multiple streams concurrently, and how to develop a scalable, robust machine learning model. 

In addition, I will also focus on the privacy and security aspects of such scalable approaches and conclude with my ongoing works and future research directions.