tarek hesham

Thesis work in progress on Geometric Deep Learning, applying mathematical structure such as graphs, manifolds, and symmetry to learning systems.

Built a custom hierarchical Vision Transformer encoder with a U-Net decoder for semantic segmentation on the CamVid road-scene dataset, and benchmarked it against a pretrained ResNet34-U-Net baseline. Deployed an interactive web demo for the trained model using Streamlit.

Created an end-to-end NLP pipeline on 21,000+ Amazon reviews to surface latent complaint themes using TF-IDF and dimensionality reduction. Compared clustering methods (K-Means, Agglomerative Clustering, DBSCAN) and learned a Bayesian Network to model dependencies behind customer ratings.

Developed a full pipeline for the BUSI dataset to train YOLOv8s for tumor detection and localization, including conversion of segmentation masks into YOLO-format bounding-box labels. Added IoU-based evaluation utilities and a lightweight MLP meta-model to refine predicted boxes and improve localization precision.

Built an end-to-end system to predict finger-joint angles from EMG signals and stream real-time predictions to a Blender-based 3D hand simulation via socket communication. Implemented an online fine-tuning feedback loop for continuous model adaptation.

Designed a hybrid CNN–Transformer architecture for fine-grained food image classification across 251 categories trained from scratch under limited compute. Improved validation accuracy by combining convolutional local features with global self-attention and added a rotation-prediction self-supervised pretraining task.

Implemented a Transformer architecture from first principles (custom self-attention and multi-head attention) for NLP-based sentiment classification.

Activities and societies: Thesis direction (in progress): Geometric Deep Learning using graphs, manifolds, and symmetry.

M.Sc. student in Artificial Intelligence for Science and Technology, with current thesis work in Geometric Deep Learning. Coursework includes foundations in mathematics, physics, and statistics, plus computer vision and deep neural network architectures.

Grade: Excellent

B.Sc. in Electronics and Communications Engineering, completed with an Excellent grade. Built a strong technical foundation supporting later work in machine learning and computer vision.