Over the next ten years, computer science graduates are poised for extraordinary opportunities, particularly as the baby-boomer generation begins their retirement phase. Acadia University, which consistently ranks in the top three for undergraduate studies in Canada, offers superior degree programs in computer science. Our students receive a comprehensive education in computer science and their interpersonal and team skills are honed through project-oriented courses and cooperative education options.

We invite you to explore our website, investigate the various degrees and certificate programs, our areas of research, funding opportunities, and also our cooperative education program. Most importantly, get to know our faculty and students through the many resources on the website. From this we think you'll discover why our motto is "Come as a student, leave as a colleague".

 

REAL-TIME REM SLEEP STAGE CLASSIFICATION USING DEEP CONVOLUTIONAL NEURAL NETWORKS

MSc Computer Science Candidate: Ahmad Chowdhury

6 December 2024

Hybrid Model Defence

Acadia Location:
CARN 410

Virtual Option:
MS Teams
Join the meeting now
Meeting ID: 265 253 234 119
Passcode: Fq6Ye7jN

10:00 AM Atlantic

Thesis Committee:

Drs. Daniel Silver & Sazia Mahfuz, Supervisors
Dr. Jiju Poovvancheri, SMU, External Examiner
Dr. Amir Eaman, Internal Examiner
Dr. Darcy Benoit, Director of the School of Computer Science
Dr. Richard Karsten, Chair of the defence

Abstract

We provide a comparative study of convolutional, recurrent, and transformer neural networks for the classification of REM sleep stages in real-time using data from a commercially available smartwatch. We have developed software for the iPhone and Apple Watch that enables real-time sensor data collection. This information is used twofold: (1) it is stored for the development and evaluation of the neural network models, (2) it is used as input to an embedded model in an iPhone app which classifies the current sleep stage in real time. Heart rate and acceleration measures are collected and consolidated into 2-minute intervals, for input to the model. Ground truth data has been gathered through Apple’s HealthKit dataset. We statistically showed that Apple’s HealthKit data was sufficiently close to the industry standard Cerebra Sleep Study System and more affordable, so going forward with our experiment, we stuck with HealthKit data for ground truth. We collaborated with the psychology department of Acadia University to conduct a sleep study with three different participants to collect data for our research. Earlier research using data from several subjects determined that the variability between person’s sleep patterns requires every person to have a customized model. To improve model performance, we used a 5-fold cross-validation to evaluate neural network architectures tailored to individual users. By integrating advanced features and incorporating Multi-Task Learning (MTL) techniques, we achieved significant improvements in real-time REM sleep detection, exceeding the benchmarks set by prior study. Our results indicate that the single-layer convolution model with batch normalization and a kernel size of 3 (CNNKernel3), as well as the single-layer convolution model with max-pooling, and kernel size of 9, had superior performance. The MTL CNNKernel3 model, derived from a single subject, scored an AUC of 0.901, correlation of 0.697, REM accuracy of 0.853, F1 score of 0.708, and TPR of 0.788 with a TNR of 0.881 on test datasets.

 

About Ahmad…

My name is Ahmad, and I am from Bangladesh, currently pursuing a master’s degree in computer science while working as a graduate research and teaching assistant. As a research assistant, I worked with a startup to develop their first minimum viable product (MVP), which aligns with my thesis on "Real-time REM Sleep Stage Classification using Deep Convolutional Neural Networks.

Additionally, in my first year, I completed two data mining and data science projects, both accepted at international conferences, with one already published in a scientific journal. These achievements reflect my dedication to advancing the field of computer science through impactful research. After completing my thesis, I plan to continue refining my thesis project to prepare it for real-world applications.

 

A Comparative Study of Generative AI and Traditional Machine Learning for Cyber-Attack Detection in VANETs

MSc Computer Science Candidate: Sony Guntuka

6 December 2024
MS Teams
Join the meeting now
Meeting ID: 230 728 125 386
Passcode: cEitrT

10:00 AM Atlantic

Thesis Committee:

Dr. Elhadi Shakshuki, Supervisor
Dr. Haroon Malik, Marshall Univ., External Examiner
Dr. Martin Tango, Internal Examiner
Dr. Darcy Benoit, Director, School of Computer Science
Dr. Zoë Migicovsky, Chair of the defence

Abstract

Vehicular Ad Hoc Networks (VANETs) are considered extremely important section of intelligent transportation system because it enables communication between infrastructure and vehicles for enhancing road safety, and efficiency. As the connectivity level increases between vehicle and infrastructure, then networks are also exposed to growing number of sophisticated cyberattacks particularly zero-day attacks that pose significant threat. The use of Traditional Machine Learning (ML) models can only detect known cybersecurity attacks and face problems in identifying unseen and novel threats. Based on this, the research will explore in detail about the potential of GenAI models like variational autoencoders, generative adversarial networks, and diffusion models for detecting unseen cyberattacks in vehicular networks. A detail comparison of GenAI models and traditional ML was conducted by focusing on detection rates, accuracy, robustness to zero-day attacks, and false positives and negatives. The results showed that Hybrid models by combining ML and GenAI outperformed standalone approaches and achieve high resilience and accuracy. Besides vital advantages of GenAI some challenges also discussed in detail like data availability, computational complexity, and adversarial vulnerability. The study concludes through outlining the contribution, limitations, and future research work for enhancing cybersecurity in VANETs through GenAI models.

 

About Sony …

I am a master's student in the Computer Science program at Acadia University, with a primary research focus on Vehicular Adhoc Networks (VANETs). Since beginning my graduate studies, I have explored various topics within this field, publishing multiple research papers on VANETs as well as related areas like IoT and Cloud Computing. My recent work delves into the potential applications of Generative AI (GenAI) for enhancing security in Vehicular Networks. Through my latest research paper, I identified promising ways in which GenAI could aid in detecting cyber-attacks within Vehicular Networks, especially given the sensitivity of the data being transmitted. This insight has led me to dedicate my thesis to developing methods for detecting unseen cyber threats in Vehicular Networks, aiming to contribute to the security and resilience of these emerging systems.