On-device AI
Running deep neural networks on resource-constrained mobile devices presents significant challenges in latency, accuracy, and hardware utilization. We develop system techniques across the model, runtime, and application stacks, including dynamic reconfiguration of vision transformers through neural stitching [EMSOFT '26], optimized vision transformer inference on mobile NPUs [EuroSys '26], and coordinated use of heterogeneous mobile processors for real-time vision foundation model inference in augmented reality [MobiSys '25].
Mobile immersive computing
Immersive media such as omnidirectional and volumetric video can deliver interactive and engaging mobile experiences, but their large data sizes and computational demands create substantial systems challenges. We develop techniques that maximize visual quality under strict time, bandwidth, and resource constraints, including super-resolution-enhanced live 360-degree video streaming [MobiSys '23], energy-optimized super-resolution for mobile devices [MobiCom '25], and fully immersive volumetric video streaming based on 3D Gaussian splatting [MobiCom '25].
Energy-aware mobile systems
Reducing energy consumption is a persistent challenge for mobile systems. We investigate system-level energy optimization across mobile native applications [MobiSys '19], mobile web browsing [MobiCom '19, EMSOFT '25], and mobile games [TMC '22]. We also design energy-efficient on-device AI techniques for ultra-low-power sensing [TECS '25] and computation-intensive mobile vision workloads such as super-resolution [MobiCom '25].
Cross-device computing
As users increasingly interact with multiple personal devices, cross-device computing has become an important part of the mobile experience. Our research pursues platform-agnostic methods by leveraging the web as a common meta-platform. We have examined how users experience the mobile web [INFOCOM '21] and developed techniques for sharing device I/O across web applications [INFOCOM '23] and distributing graphical interfaces across devices [INFOCOM '24].