I am a dedicated researcher with a strong background in deep learning and computer vision. I completed both my Bachelor's degree in Electrical Engineering and my Master's degree in Computer Science from the National University of Sciences and Technology (NUST) in Islamabad, Pakistan. Throughout my academic journey, I developed a passion for solving complex problems using advanced machine learning techniques, particularly in the areas of image processing and computer vision.
Currently, I am pursuing my PhD at Mälardalens University in Västerås, Sweden. My research, under the umbrella of the AutoFL project, focuses on the intersection of Federated Learning (FL) and Neural Architecture Search (NAS). I am deeply interested in developing methodologies that leverage deep learning models within a federated learning setup, enabling multiple clients or devices to jointly train these models while meticulously maintaining data privacy and robustness. A key aspect of my work involves applying Neural Architecture Search to automatically discover optimal model architectures tailored for these decentralized, privacy-preserving FL environments.
Before embarking on my PhD, I led the AI team at the Machine Vision and Intelligent Systems (MachVIS) Lab at NUST, where I guided numerous projects in AI and deep learning. My experience also includes working as a Machine Learning Engineer at the National Engineering and Scientific Commission (NESCOM), where I developed and deployed deep learning models for image matching and object detection tasks. I am proficient in Python, C, C++, and SQL, and have extensive experience with leading frameworks such as PyTorch, Keras, and OpenCV, which has equipped me with the versatility and skills needed to excel in the fields of machine learning and computer vision.
My research interests lie at the intersection of Machine Learning and Computer Vision, with a particular focus on Federated Learning (FL) and Neural Architecture Search (NAS). I have been working extensively on developing methodologies that allow deep learning models to be efficiently designed and collaboratively trained in decentralized environments. This work addresses the critical challenges of data privacy, model robustness, and computational efficiency inherent in distributed machine learning systems.
Currently, I am involved in the AutoFL project, where I am exploring the challenges and opportunities of implementing Neural Architecture Search within a federated learning setup. This involves designing automatic methods to discover optimal model architectures that can be effectively trained across multiple, independent clients or devices. My goal is to leverage the strengths of distributed learning and automated model design to enhance the performance and applicability of deep learning systems in diverse and heterogeneous environments, pushing the boundaries of what is possible with privacy-preserving, collaborative AI.
FedLoRASwitch: Efficient Federated Learning via LoRA Expert Hotswapping and Routing (Oct 2025) Joakim Flink, Bostan Khan, Masoud Daneshtalab The 3rd IEEE International Conference on Federated Learning Technologies and Applications (FLTA25)
HeRD: Modeling Heterogeneous Degradations for Federated Super-Resolution in Satellite Imagery (Jul 2025) Bostan Khan, Seyedhamidreza Mousavi, Masoud Daneshtalab Journal of IEEE Access (IEEE-Access)