Nauman Ali Murad

The emergence of agentic Artificial Intelligence (AI), which can operate autonomously, demonstrate goal-directed behavior, and adaptively learn, indicates the onset of a massive change in today's computing infrastructure. This study investigates how agentic AI models' multiple characteristics may impact the architecture, governance, and operation under which computing environments function. Agentic AI has the potential to reduce reliance on extremely large (public) cloud environments due to resource efficiency, especially with processing and/or storage. The aforementioned characteristics provide us with an opportunity to canvas the likelihood of strategic migration in computing infrastructures away from massive public cloud services, towards more locally distributed architectures: edge computing and on-premises computing infrastructures. Many of these likely migrations will be spurred by factors like on-premises processing needs, diminished data consumption footprints, and cost savings. This study examines how a solution for implementing AI's autonomy could result in a re-architecture of the systems and model a departure from today's governance models to help us manage these increasingly autonomous agents, and an operational overhaul of processes over a very diverse computing systems landscape that bring together computing via cloud, edge, and on-premises computing solutions. To enable us to explore these intertwined decisions, it will be fundamentally important to understand how to best position agentic AI, and to navigate the future state of computing infrastructures.

Glacial Lake Outburst Floods (GLOFs) pose a severe risk to populations in high-altitude areas, particularly in Pakistan's northern regions, where glacier melt has created 3,044 glacial lakes in Gilgit-Baltistan and Khyber Pakhtunkhwa. The Chitral region is especially vulnerable, highlighting the need for robust monitoring and mitigation strategies. This paper focuses on using Deep Learning (DL) models for detecting and segmenting glacial lakes to mitigate GLOF risks. Using the Glacial Lakes Detection Dataset from High-Mountain Asia, covering 2080.12 km² and including around 30,121 glacial lakes, we selected 1,200 cloud-free true-color images with ground truth masks. Our study evaluates DL models like DeepLabV3+, U-Net, and YOLOv8 for lake segmentation and classification. Experimental results demonstrated model proficiency, with DeepLabV3+ (ResNet50 backbone and Dice loss function) achieving the highest IoU score of 77.2%. These findings highlight the potential of DL-based systems for enhanced GLOF monitoring and early warning.