Saeed Ahmad
About Me
AI Researcher
I am an AI Researcher specializing in knowledge distillation, multimodal learning, and efficient deep learning for medical imaging and embedded systems. My research focuses on developing novel techniques to compress and optimize deep neural networks while maintaining state-of-the-art performance.
Currently, I work as an AI Research Engineer at IK Lab Inc. in Seoul, where I develop and deploy custom AI models on embedded devices including NVIDIA Jetson and Raspberry Pi with Hailo accelerators. My work spans atmospheric turbulence mitigation, infrared-visible image fusion, and real-time object detection systems.
I completed my Master's degree in Artificial Intelligence at Korea National University of Transportation under the supervision of Professor Jeonghwan Gwak, where I researched multi-teacher cross-modal knowledge distillation for medical image segmentation. My work has been published in high-impact journals including Knowledge-Based Systems (IF: 7.2).
Beyond research, I have extensive industry experience with 50+ freelance projects in machine learning and computer vision, demonstrating my ability to translate research into practical applications.
Publications
Peer-reviewed research in knowledge distillation, medical imaging, and efficient neural networks
Multi-teacher cross-modal distillation with cooperative deep supervision fusion learning for unimodal segmentation
Implementing Cost-Effective CNNs through INT8 Quantization Aware Training on Embedded Systems
Efficient Dynamic Pruning Technique for Reduction of Inference Time in Convolutional Neural Network
Research Projects
Ongoing and completed research initiatives
Atmospheric Turbulence Mitigation for Embedded Systems
Developed and optimized deep learning models for real-time atmospheric turbulence mitigation in images and videos, deployed on NVIDIA Jetson devices using TensorRT and DeepStream.
Continual Learning with KAN-based Vision Transformers
Implemented continual learning with Kolmogorov-Arnold Networks based Vision Transformers, enabling models to learn new tasks while retaining knowledge from previous ones and mitigating catastrophic forgetting.
Anomaly Detection using Normalizing Flows
Developed unsupervised anomaly detection and localization system for industrial heat radiators using normalizing flows, with containerized deployment via Docker and Flask.
Infrared-Visible Image Fusion for Real-Time Monitoring
Built multimodal image fusion systems that combine infrared and visible spectrum imagery for enhanced real-time monitoring applications on embedded devices.
Multi-Task Whisper for Korean Kids' Speech Recognition
Fine-tuned OpenAI's Whisper model for Korean language with auxiliary heads for age, gender, and dialect classification, achieving state-of-the-art performance on kids' speech datasets.
Research Interests
Areas of active research and expertise
Knowledge Distillation
Compressing large models into efficient student networks while preserving performance
Medical Image Analysis
Deep learning for segmentation, classification, and detection in clinical imaging
Model Optimization
Quantization, pruning, and efficient inference for edge deployment
Multimodal Learning
Fusing information from multiple modalities for robust predictions
Low-Level Vision
Image restoration, super-resolution, denoising, and enhancement
Continual Learning
Enabling models to learn new tasks while retaining knowledge from previous ones
Experience
Research and professional journey
AI Research Engineer
IK Lab Inc., Seoul, South Korea
Developing and deploying custom AI models on embedded devices including NVIDIA Jetson Series and Raspberry Pi with Hailo Accelerator.
- Atmospheric turbulence mitigation in images and videos
- Infrared and visible image fusion for real-time monitoring
- Object detection systems using NVIDIA DeepStream with TensorRT optimization
Research Assistant
AMI Lab, Korea National University of Transportation
Conducted research in medical image analysis and efficient deep learning under Professor Jeonghwan Gwak.
- Designed multi-teacher cross-modal knowledge distillation for unimodal segmentation
- Implemented INT8 quantization-aware training on embedded systems
- Developed continual learning with KAN-based Vision Transformers
- Built unsupervised anomaly detection using normalizing flows
Machine Learning Engineer
DiveDeepAI, Islamabad, Pakistan
Built machine learning solutions for financial technology applications.
- Developed generalized ML pipeline for stock market forecasting
- Implemented rule-based intelligent expert engine for market analysis
Education
Academic background and qualifications
Master of Artificial Intelligence
Korea National University of Transportation, South Korea
CGPA: 4.38 / 4.50
Thesis: Advancing Medical Image Segmentation Using Cross-Modal Knowledge Distillation and Hybrid CNN-Transformer
Supervisor: Professor Jeonghwan Gwak
Scholarship: Recipient of Professor-based Scholarship (NRF funded projects)
Bachelor of Computer Science
Pakistan Institute of Engineering and Applied Science (PIEAS), Islamabad
CGPA: 3.89 / 4.00
Thesis: Machine Learning Based Network Intrusion Detection System
Supervisor: Dr. Muhammad Abid Mughal
Activities: President of PIEAS Cyber Security Club | ML/AI Lead at Google DSC PIEAS
Applied ML Experience
Bridging research and real-world applications through freelance and open-source work
Beyond research, I have extensive experience applying machine learning to solve real-world problems through freelance consulting and open-source contributions. This practical experience informs my research by exposing me to diverse domains and constraints.
Client Testimonials
Saeed is very helpful and has great skills. The model he made works very well and I am glad I chose him.
Client from Italy
Good experience with Saeed. He solved the problem of medical image segmentation in a very short time. Highly professional.
Client from United Kingdom
Saeed was able to build a multi-class deep learning model for retinal disease classification with over 28 classes. Great job.
Client from UAE
Completed an autoencoder and GAN task in TensorFlow. Delivered one day before the deadline. Quality work.
Client from Sri Lanka
View my Fiverr profile for more reviews and service details.
Open Source Contributions
Python packages and tools shared with the community
Teeth Segmentation
Deep learning-based teeth segmentation from dental images. A complete pipeline for automated dental image analysis.
YOLOv12 TensorRT C++ with CUDA NMS
High-performance YOLOv12 inference implementation with TensorRT optimization and GPU-accelerated Non-Maximum Suppression in C++.
Get in Touch
Open to research collaborations, consulting, and opportunities
Location
Seoul, South Korea
saeedahmad.icp@gmail.com
Phone
+82-10-6672-0250