MES Lab @ Ewha

On-Device AI

Mobile & Edge AI

We develop resource-efficient AI systems that enable advanced DNNs, vision foundation models, and perception pipelines to run directly on mobile and edge devices. Our work adapts model execution to device conditions such as latency, energy, memory, temperature, and heterogeneous processor availability.

Representative Works

EMSOFT 2026DynaStitch: Dynamic ViT Inference via Neural Stitching for Resource-Constrained Edge Devices
TECS 2026Phoenix: Thermal-Aware On-Device Inference of Multi-Instance DNNs for Mobile Video Applications
MobiSys 2025ARIA: Optimizing Vision Foundation Model Inference on Heterogeneous Mobile Processors for Augmented Reality
MobiCom 2024Panopticus: Omnidirectional 3D Object Detection on Resource-constrained Edge Devices

Energy-Aware Mobile and Embedded Systems

Mobile Systems

We design mobile and embedded systems that understand, predict, and optimize energy use in real-world applications. Our work targets practical workloads such as mobile video, web browsing, smart mobility, and safety-critical sensing, where energy efficiency must be balanced with performance, reliability, and user experience.

Representative Works

Intelligent Energy Storage and Battery-Less IoT

Energy Systems

We build intelligent energy systems for batteries, hybrid storage platforms, energy harvesters, and battery-less IoT devices. Our research combines machine learning, hardware design, sensing, and system optimization to improve reliability and autonomy under dynamic energy availability.