01
Extended Reality Systems
End-to-end systems for immersive media: from volumetric video pipelines to latency-aware delivery and interaction.
Research
Research areas.
We build and measure systems under the constraints they actually run in: variable networks, limited devices, and latency budgets that users notice when you miss them.
02
AI at the Edge
Practical AI systems that fit the constraints of devices, accelerators, networks, and the people using them.
- IEEE SECON 2026DualEngine: A Thermal-Aware Vision Inference Framework via Mobile and Cloud Co-Execution
- IEEE Transactions on Mobile Computing 2025NeuroBalancer: Balancing System Frequencies with Punctual Laziness for Timely and Energy-efficient DNN Inferences
- ACM MobiSys 2024CoActo: CoActive Neural Network Inference Offloading with Fine-grained and Concurrent Execution
- ACM Multimedia 2023ENTRO: Tackling the Encoding and Networking Trade-off in Offloaded Video Analytics
- ACM GetMobile 2022zTT: Learning-based DVFS with Zero Thermal Throttling for Mobile Devices
- ACM MobiSys 2021zTT: Learning-based DVFS with Zero Thermal Throttling for Mobile Devices
03
Multi-Agent Systems
We design collaborative AI agents that reason, coordinate, and act together to solve complex tasks across real systems.
04
Networked Systems
Networks that adapt to the traffic they carry: cross-layer scheduling, transport, and measurement for real-time and learning-driven workloads.
- ACM SIGCOMM 2026DeepSFU: Scalable Deepfake Detection for Video Conferencing
- IEEE INFOCOM 2026PAVE: Mitigating Non-Congestive Delay for Seamless Video Calls over NextG Mobile Networks
- USENIX NSDI 2026eXpressSFU: Toward Super-Scalable Video Conferencing with SmartNICs
- USENIX NSDI 2026QCON: Seamless QoE-Aware 5G Streaming via Multi-Connectivity
- IEEE DySPAN Workshop 2024Repurposing Cellular Reference Signals: Accurate RSRP Measurements with Mobile Phones
- ACM Multimedia 2023ENTRO: Tackling the Encoding and Networking Trade-off in Offloaded Video Analytics
- ACM Multimedia 2022RL-based FEC Adjustment for Better QoE in WebRTC
05
AI Infrastructure
Scheduling, memory, and resource management for the clusters that train and serve models — where the workload itself is learned.
How we work
Go deep into the system.
That is where the wins are.
- 01 Start from a bottleneck we have measured, not assumed.
- 02 Take the idea all the way to an end-to-end implementation.
- 03 Evaluate on real devices and real workloads.