-
Research Intern at University of Alberta
Mitacs GRI Internship (06/2023 - 09/2023)
Project Title : Exploring Machine learning and Monte Carlo
Algorithms for Wireless Sensor Networks with
Energy Replenishment
Project Supervisor : Dr. Ehab Elmallah
- Conducted research on energy-harvesting wireless sensor networks (EHWSNs).
- Developed and evaluated node energy control methods for day-to-day operation using discrete-event simulation (DES) programming.
- Implemented routing functionality and documented the simulator program (using OMNET++) for EHWSNs.
- Designed and executed simulation experiments to assess performance and energy control strategies, including those with machine learning techniques.
- Documented design aspects and discussed results with an emphasis on enhancing EHWSN performance.
-
Research Assistant at Minnesota State University
Global UGRAD Pakistan (01/2023 - 05/2023)
Project Title : Efficient Anomaly Detection in IIoT Networks
Project Supervisor : Dr. Naseef Mansoor
- Developed a solution to enhance the security of IIoT devices, addressing the growing challenges associated with their prevalence.
- Focused on the critical aspect of anomaly detection in IIoT networks, aiming to overcome issues like high false positives and low sensitivity encountered by traditional methods.
- Pioneered a practical, cost-effective, and low-latency anomaly detection technique, underscoring the innovative nature of the approach.
- Utilized a cutting-edge pre-trained global model based on the gated recurrent unit (GRU) algorithm and implemented it locally on each router through federated learning.
- Leveraged the unique behavior of individual device types, achieving an impressive 87 percent accuracy in anomaly detection, all while preserving communication efficiency and data privacy.
- Demonstrated the effectiveness of the approach through a comprehensive tradeoff analysis involving key metrics such as area, time, and accuracy, showcasing its superiority over conventional methods.
-
Research Assistant at AEYE Health
Bachelor’s Thesis (05/2022 - Present)
Project Title : Diabetic Retinopathy Diagnostics using
Federated Learning Approach
Project Supervisor : Dr. Saad Ahmed Qazi
- Proposed and implemented a federated learning approach to address data privacy concerns in medical imaging for Diabetic Retinopathy.
- Prepared and preprocessed datasets for federated learning, considering both IID and non-IID characteristics.
- Conducted experiments with various federated learning strategies, including FedAvg, FedProx, and SCAFFOLD, resulting in impressive accuracy rates of 85.79%, 87.9%, and 88.02%, respectively.
- Explored the integration of Differential Privacy in a cross-silo federated setup to enhance privacy in machine learning.
- Performed a comprehensive analysis of privacy-preserving techniques, assessing options like encrypting model weights, involving trusted third parties, and employing trusted execution third parties.
- Implemented Differential Privacy (DP) with the addition of Gaussian noise to measure privacy leakage and evaluate privacy budget consumption.
- Showcased the effectiveness of the mobilev2net model in diabetic retinopathy prediction, achieving an accuracy of 85.79% without noise and 82.04% with a noise multiplier of 8.0.
-
Research Assistant at NEUROCOMPUTATIONAL LAB
(05/2022 - 12/2022)
Project Title : Real-time Personal Protective Equipment Detection in Healthcare
- Developed real time detector which can detect if people are wearning PPE such as face mask, face shield, gloves, and gown.A hardware pipeline is created using Raspberry Pi, HD Webcam, PIR sennsor, and OptoCoupler Relay Module for PPE detection.
- On PIR motion sensor detection, the application opens a front camera preview, collect all the frames, detect faces present in the frames, crop down the faces and converts them to bitmaps.
- Preprocessing is done on images and a combined dataset is prepared consisting of collected images and our captured images.
- Created YOLOv4 computer vision model which specially performs well in real time object detection.
- Annotation and augmentation is done on all the training set images.
- The detector weight is also converted to TensorFlow format to check live detection performance and added features like live object count and record keeping.
-
Cyber Security Intern at NATIONAL CENTER FOR CYBER SECURITY
(01/2021 - 07/2021)
Project Title : Endpoint Detection and Response
- Used Endpoint Detection and Response (EDR) capabilities using Osquery to query endpoints and Fleet as a control server for Osquery.
- Wrote Kolide API and Websocket portion of the API with Python in Fleetdm source code, to retrieve live query results.
- Stored live query results in Redis database which is then forwarded to the Elastic Stack for monitoring.
- Implemented Random Forest algorithm to detect TrickBot malware infections flows without having to analyse network packet payloads, the IP addresses, port numbers and protocol information.
- Based on the analysis it shows that the Random Forest classifier identifies TrickBot related flows with 99.9534% accuracy, 91.7% true positive rate.
-
Web Development Intern at RESEARCH CENTER FOR ARTIFICIAL INTELLIGENCE
(08/2021 - 01/2022)
- Created RCAI webpage.
- The webpage is made out of AngularJS with UI elements from Materialize.
- Used Django to connect with SQL backend with SQL Server 2012.