Sayed Erfan Arefin

Assistant Professor, Computer Science

University of Dayton

Dayton, OH 45459

I am an Assistant Professor in the Department of Computer Science at the University of Dayton. My research focuses on cybersecurity, with particular interests in deep learning, side-channel attacks, large language models, Internet of Things (IoT) network architectures, and robotics security.

I completed my Ph.D. at Texas Tech University, where my dissertation was titled “Covert Observation of Deep Neural Network Operations for Architectural Inference Attacks.” This work delved into architectural inference attacks on deep neural network using side-channels, contributing to the field’s understanding of potential vulnerabilities in AI systems.

Prior to my doctoral studies, I earned both my Bachelor’s and Master’s degrees in Computer Science and Engineering from BRAC University in Dhaka, Bangladesh. These formative years solidified my passion for computer science and laid the groundwork for my future research endeavors.

I am always open to collaborations and excited to explore new research opportunities. If you’re interested in working together or learning more about my work, please feel free to reach out.

selected publications

Music to My Ears: Turning GPU Sounds into Intellectual Property Gold

Arefin, Sayed Erfan, and Serwadda, Abdul

In Proceedings of the 2024 Workshop on Artificial Intelligence and Security 2024

Abs PDF Poster

In this paper, we introduce an acoustic side-channel attack that extracts crucial information from Deep Neural Networks (DNNs) operating on GPUs. Utilizing a Micro-Electro-Mechanical Systems (MEMS) microphone with an extensive frequency range, we demonstrate that the distinct sounds produced during DNN operations can inadvertently reveal significant details about the network’s architecture. Through extensive experimentation with a variety of neural networks from the ImageNet competition, we validate the efficacy of this novel attack vector. Our contributions include a detailed methodological framework for capturing and analyzing acoustic data, empirical validation using prominent ImageNet models, and a comprehensive sensitivity analysis to assess the attack’s robustness under varying conditions. This research not only uncovers a previously unexplored vulnerability in neural network security but also provides a foundation for developing more robust defense mechanisms against such innovative side-channel attacks.
IoT Infrastructure: Fog Computing Surpasses Cloud Computing

Ashrafi, Tasnia H., Hossain, Md. Arshad, Arefin, Sayed E., Das, Kowshik D. J., and Chakrabarty, Amitabha

In Intelligent Communication and Computational Technologies 2018

Abs

The Internet of Things can be defined as an expansion of Internet network further off conventional gadgets like computers, smart phones, and tablets to an assorted scope of gadgets and ordinary things that use embedded technology to communicate with the outer environment, all by means of the Internet. With the increasing time, data access and computing are proceeding towards more complications and hurdles requiring more efficient and logical data computation infrastructures but the vision of future IoT seems foggy as it is a matter of concern that our current infrastructure may not be able to handle large amount of data efficiently involving growing number of IoT devices. To get diminished from this issue, Fog computing has been introduced which is an expansion of cloud computing working as a decentralized infrastructure in which reserving or computing data, service, and applications are distributed in the most efficient and logical position between source and the cloud. In this paper, the current infrastructure has been depicted and proposed another model of IoT infrastructure to surpass the difficulties of the existing infrastructure, which will be a coordinated effort of Fog computing amalgamation with Machine-to-Machine (M2M) intelligent communication protocol followed by incorporation of Service-Oriented Architecture (SOA) and finally integration of agent-based SOA. This model will have the capacity to exchange data by breaking down dependably and methodically with low latency, less bandwidth, heterogeneity in less measure of time maintaining the Quality of Service (QoS) precisely.
Service Based FOG Computing Model for IoT

Ashrafi, Tasnia Heya, Hossain, Md. Arshad, Arefin, Sayed Erfan, Das, Kowshik Dipta Jay, and Chakrabarty, Amitabha

In 2017 IEEE 3rd International Conference on Collaboration and Internet Computing (CIC) Oct 2017

Abs

The more we are heading towards the future with ever-growing number of IoT devices which is expected to take place in a giant number almost near trillions by 2020, the data access and computing are proceeding towards more complications and impediments requiring more efficient and logical data computation infrastructures. Cloud computing is a centralized Internet based computing model which acts like a storage as well as a network connection bridge between end devices and servers. Cloud computing has been ruling as a data computation model for quite a while but if we try to concentrate on the upcoming IoT generation, the vision will become a little bit blurry as it is not possible for the present cloud computing models to deal with such huge amount of data and to rescue from this foggy situation, Fog computing model comes forward. Instead of being a replacement of the cloud computing model, Fog computing model is an extension of Cloud which works as a distributed decentralized computing infrastructure in which data compute, storage and applications are distributed in the most logical, efficient place between the data source and the cloud. In this paper, we proposed an infrastructure with collaboration of Fog computing combined with Machine-to-Machine(M2M) intelligent communication protocol followed by integration of the Service Oriented Architecture(SOA) and this model will be able to transfer data by analyzing reliably and systematically with low latency, less bandwidth, heterogeneity in less amount of time maintaining the Quality of Service(QoS) befittingly.
Image Processing Based Indoor Localization System for Assisting Visually Impaired People

Heya, Tasnia Ashrafi, Arefin, Sayed Erfan, Chakrabarty, Amitabha, and Alam, Mazidul

In 2018 Ubiquitous Positioning, Indoor Navigation and Location-Based Services (UPINLBS) Mar 2018

Abs

Indoor localization or indoor positioning system is a known as a process of detecting position of any object or people inside a building or room by different sensory data collected from different devices using different techniques such as radio waves, magnetic fields, acoustic signals or other procedures. However, lacking of a standard localization system is still a very big concern. Solution of this issue can be very beneficial for people in many cases but it can be especially very beneficial for the visually impaired people. In this paper, an image processing based indoor localization system has been developed using OpenCV and Python by following color detection technique to detect position of the user with maximum accuracy and then location of user is determined by analyzing that location matrix. Location accuracy depends on the size of the matrix and successful identification of target color. Firebase real time database was added to the system which made real time operations between server and the user end device easier. To justify the proposed model, successful experiments were conducted in indoor environments as well and correct result was achieved each time by detecting accurate locations. This will be very advantageous to observe the fully or partially sightless people and guide them towards their destination and also to inspect them for their security purpose.
Deep Neural Exposure: You Can Run, But Not Hide Your Neural Network Architecture!

Arefin, Sayed Erfan, and Serwadda, Abdul

In Proceedings of the 2021 ACM Workshop on Information Hiding and Multimedia Security Mar 2021

Abs PDF Slides

Deep Neural Networks (DNNs) are at the heart of many of today’s most innovative technologies. With companies investing lots of resources to design, build and optimize these networks for their custom products, DNNs are now integral to many companies’ tightly guarded Intellectual Property. As is the case for every high-value product, one can expect bad actors to increasingly design techniques aimed to uncover the architectural designs of proprietary DNNs. This paper investigates if the power draw patterns of a GPU on which a DNN runs could be leveraged to glean key details of its design architecture. Based on ten of the most well-known Convolutional Neural Network (CNN) architectures, we study this line of attack under varying assumptions about the kind of data available to the attacker. We show the attack to be highly effective, attaining an accuracy in the 80 percentage range for the best performing attack scenario.