Wael Sulais

تخصص الذكاء الاصطناعي · دفعة 2025

Automatic Soccer Penalty Detection (Trajectory of Ball and Goalkeeper)

Detecting the trajectory of the soccer ball and goal keeper with their pose to predict whether the shot would cause a goal. From the collected data we can get statistics such as “Expected Goal on Target (xGOT)”. Furthermore, you can use similarity search from the detections to find similar penalty shots.

اقرأ التقرير النهائي (PDF)

اطّلع عليه في تقرير البرنامج (PDF)

تخصص المعلوماتية الحيوية · دفعة 2024

YOLO-AR: A CT Process Analytics Framework

Diagnostic medical imaging services require significant resources, including equipment, materials and labor. Proper coordination of these resources positively impacts the productivity of the radiology department, which, in turn, influences access to care, costs, and quality. Thus, the availability of near-real-time productivity measurements may improve department operations by providing insight into system bottlenecks. We propose a novel Operations Research (OR) framework for measuring the efficiency of a hospital's medical imaging department. Utilizing a mix of AI methods such as image segmentation and action recognition, we provide a state-of-the-art (SOTA) system for tracking the end-to-end process latency and patient inventory while preserving privacy. More specifically, we explore YOLOv8 for image segmentation as well as ResNet18 for action recognition. As a part of this endevor, we have collected footage of simulated patient computer tomography (CT) scans for training. Furthermore, the system has been deployed on a Nvidia Jetson Orin NX in a teaching hospital. Henceforth, We believe that our system can be applied to more hospitals in Greater Toronto Area (GTA), generalized to more types of medical imaging procedures, as well as extended to produce more fine-grained analytics.

اقرأ التقرير النهائي (PDF)

اطّلع عليه في تقرير البرنامج (PDF)

تخصص الذكاء الاصطناعي · دفعة 2023

SABBAH: Multi-Modal Arabic Hate Speech Detection

The SABBAH project is a pipeline of models that are designed to separate voices and classify hate speech. The project uses a combination of Vocal Seperation Models, Automatic Speech Recognition, and Arabic LLMs to identify hate speech from text and audio recordings. Cutting edge techniques such as adversarial training have been applied to improve the models reslience to malicious inputs. The SABBAH project is still under development, but it has the potential to make a significant impact on the field of AI safety

اطّلع عليه في تقرير البرنامج (PDF)