IIIT Hyderabad students develop dynamic ML model-switching approach for traffic monitoring
Hyderabad: A team of second-year Computer Science and Engineering (CSE) students from IIIT Hyderabad has demonstrated a dynamic machine learning (ML) model-switching approach on smartphones for real-time traffic monitoring. Their research, which resulted in a paper titled “EdgeML Balancer: A Self-Adaptive Approach for Dynamic Model Switching on Resource-Constrained Devices,” has been accepted for presentation at the International Workshop on Flexible Software Architecture for Embedded Computing Systems (SARECS) 2025, held alongside the International Conference on Software Architecture (ICSA).
The team, comprising Kriti Gupta, Ananya Halgatti, Priyanshi Gupta, and Larissa Lavanya, participated in the Embedded Systems Workshop (ESW), a hands-on course that focuses on experiential learning for second-year CSE students. In search of a unique project, they discovered the Qualcomm Innovators Development Kits (QIDK), which were available through a collaboration between IIITH and Qualcomm. The kits, featuring the latest Snapdragon system-on-chip (SoC), provided the students with the opportunity to explore and experiment with edge AI technologies.
Under the guidance of Prof. Karthik Vaidyanathan from the Software Architecture 4 Sustainability (SA4S) group, and PhD student Akhila Matathammal, the students developed the EdgeML Balancer. The approach aims to optimize model-switching decisions on edge devices like smartphones based on operational contexts such as user requests, response time, accuracy, and energy consumption. The team prototyped their model on the QIDK platform and tested it with real-time traffic data.
“This work extends previous research in model-switching from the cloud to resource-constrained edge devices, incorporating energy sustainability into the model-selection process,” said Akhila Matathammal. The approach addresses key challenges in Edge AI systems, and the team’s findings will be presented at the SARECS workshop in Denmark.
The research began as an exploratory project without the intention of publication. “Our primary goal was to learn and experiment with new technologies,” said Ananya Halgatti. Despite initial challenges with the QIDK platform, the team benefited from support provided by Qualcomm and senior students with prior experience. “It was a mix of frustration and fun, but we managed to overcome the hurdles with the help of our mentors,” said Kriti Gupta.
IIIT Hyderabad undergrad team develops dynamic ML model-switching approach for real-time traffic monitoring on smartphones. Their research, “EdgeML Balancer,” will be presented at SARECS 2025! #IIITHyderabad #EdgeAI #MachineLearning #SARECS2025 pic.twitter.com/kHSHnDZtaI
— Hyderabad Mail (@Hyderabad_Mail) February 27, 2025
The team’s work has now progressed to testing the model on the Samsung S24 Ultra, following initial deployments on the Samsung Galaxy M21. The next step will involve showcasing their findings at the upcoming R&D Showcase at IIITH in March.
Prof. Vaidyanathan praised the students’ achievement, noting that the success of this project highlights the value of hands-on learning beyond traditional academic assessments. “This is a great example of how undergraduate students can make a meaningful contribution to the field, and I hope it inspires others to look beyond just grades,” he said.
