Research

Rafed works on efficient AI for constrained hardware: model pruning and quantization, FPGA and HLS acceleration, computer vision, 3D point clouds and embedded deployment.

Current project

Mitacs Globalink 2026, with Prof. Seok-Bum Ko.

XD-Prune: dependency-aware cross-domain structured pruning for YOLO26n. A structured-pruning framework that aggressively reduces model cost while holding on to robust object-detection performance beyond the visual domain it was trained on. The work measures accuracy retention across general and adverse driving conditions, and prepares compact models for embedded NCNN inference on a PYNQ-Z2.

Manuscripts

LISA: a lightweight INT8 segmentation accelerator for edge AI

Co-author. Submitted manuscript.

  • A hardware-aware, fully integer segmentation design aimed at resource-constrained edge deployment
Architecture diagram of LISA, a lightweight INT8 segmentation accelerator for edge AI.

Improving PointMLP-Lite for point-cloud FPGA deployment

Co-author. Submitted manuscript.

  • Reduces model cost and improves deployment throughput for point-cloud learning workloads

Research experience

  • Mitacs Globalink Research Intern, University of Saskatchewan, Ko Lab (2026-present). Structured pruning, cross-domain evaluation, model-recovery experiments and embedded deployment analysis for lightweight object detection.
  • Researcher, Deep Learning Lab, TUKL-NUST collaboration (2024-present). FPGA acceleration and deployment-oriented optimization of 3D point-cloud neural networks, including quantization and Vitis/Vivado workflows.
  • SoC Lab Intern and Researcher, NUST (2025-present). Hardware-aware computer vision, including compact segmentation models and INT8 accelerator-oriented design.

Education

  • B.E., Electrical Engineering, National University of Sciences and Technology (NUST), Islamabad, final year. CGPA 3.70/4.00.