Dates: 2022
Principal Investigator: Dr. Irfan Ahmad
Funded by: SDAIA-KFUPM Joint Research Center for Artificial Intelligence

Description: Over the next one to two years, my research will focus on two main directions. The first is deep learning model compression and continual learning, aimed at making large models more efficient for deployment on edge devices. Building on our current work in post-training quantization for integer-only deployment, we plan to extend this to training-based quantization combined with knowledge distillation to achieve further efficiency gains. We are also exploring how these approaches can be integrated with continual learning, with applications such as privacy-preserving energy disaggregation. The second direction is machine learning and pattern recognition for text analysis and recognition, particularly in the context of Arabic language processing. Our current work includes GAN-based textual image generation for data augmentation and research on Arabic poem meter classification. Moving forward, we plan to study different tokenization schemes, explore classification using speech signals with transfer learning, and investigate automatic classification of Arabic texts by era, as well as techniques for processing figurative speech.