I am an NCSU Master's student specializing in Artificial Intelligence (AI), with a strong interest in Reinforcement Learning (RL) and Computer Vision (CV).
I have completed multiple internships in Computer Vision and am currently working towards a Machine Learning Engineer role.
📌 LinkedIn: adrian27513
Mehreen Saeed, Adrian Chan, Anupam Mijar, Joseph Moukarzel, Georges Habchi, Carlos Younes, Amin Elias, Chau-Wai Wong, Akram Khater
Published at: Advances in Neural Information Processing Systems (NeurIPS), Sep 26, 2024
📄 arXiv | 💾 GitHub | 📂 Dataset
We present Muharaf, a dataset of 1,600+ historic handwritten Arabic page images transcribed by experts.
Each document includes polygonal coordinates for text lines and essential page elements.
This dataset enhances handwritten text recognition (HTR), particularly for cursive scripts.
It spans varied handwriting styles and document types, including letters, diaries, legal correspondences, and church records.
We describe the data acquisition pipeline, dataset features, and preliminary CNN-based baseline results.
Adrian Chan, Anupam Mijar, Mehreen Saeed, Chau-Wai Wong, Akram Khater
Under Review at: International Conference for Machine Learning (ICML)
📄 arXiv | 💾 GitHub
Handwritten Arabic text recognition (HTR) is challenging due to varied writing styles and limited dataset availability.
We propose HATFormer, a transformer-based encoder-decoder model adapted from a state-of-the-art English HTR model.
HATFormer:
- Leverages attention mechanisms to capture spatial contextual information.
- Addresses cursive character differentiation, visual decomposition, and diacritic recognition.
- Uses custom Arabic text tokenization and an optimized image processor for Arabic handwriting.
Performance Highlights:
- 8.6% Character Error Rate (CER) on the largest historical Arabic HTR dataset (51% improvement over baselines).
- 4.2% CER on the largest non-historical Arabic HTR dataset.
- Demonstrates adaptability of English HTR methods to low-resource languages with script-specific challenges.
- Primary: Python, PyTorch, Transformers, Computer Vision, Reinforcement Learning
- Machine Learning: PyTorch, TensorFlow, OpenCV, Gymnasium, Retrieval-Augmented Generation, YOLO, Tesseract, NumPy, Pandas, Scikit-learn, GPT, LLMs, Transformers, Federated Learning, Vector Databases, Reinforcement Learning
Python, Java, C, TypeScript/JavaScript, HTML5, SQL, R, MATLAB
- Development: Git, Docker, FastAPI, Microservices, Spring Boot, Angular
- Databases & Cloud: AWS, PostgreSQL, MongoDB, Milvus
- Testing & API Development: Postman
If you're interested in collaborations, research discussions, or just want to connect, feel free to reach out!
📌 LinkedIn: adrian27513