A Context-Aware Neural Network for the Abstraction and Reasoning Corpus (ARC)
Designed and implemented a novel two-stage neural network (ViT-Encoder, U-Net Decoder) to solve abstract reasoning tasks from the ARC benchmark. Pioneered a task-specific training paradigm.
Matrix Factorization-Based Recommendation System
Built a scalable movie recommendation system using Alternating Least Squares (ALS) matrix factorization, optimizing user and item biases. Implemented an interactive Django-based web application for personalized movie suggestions, deployed with Docker.
Text and Data Mining for Evaluating Research Proposals
Designed an AI-powered system to analyze and assess research proposals in alignment with Oman Vision 2040. Utilized Natural Language Processing techniques like TF-IDF and multilingual embeddings.
Adversarial Attacks Against Neural Networks
Investigated adversarial attacks on AI models, comparing FGSM and one-pixel attacks based on distortion, attack success rates, and computational efficiency on MNIST and CIFAR-10 datasets.