We measure modern Internet protocols, design intelligent multipath transport, and build reliable experimental systems for web, mobile, IoT, and UAV applications.
From observing the Internet to designing new transport mechanisms, our work follows a single loop: measure, understand, build, and evaluate.
Learning-based schedulers make Wi-Fi, cellular, and future access links work together under changing conditions.
Explore research ↗
Large-scale and in-the-wild measurements reveal where modern transport protocols help—and where they do not.
Explore research ↗
Application-aware transport for web objects, adaptive video, bulk transfer, and latency-sensitive workloads.
Explore research ↗
Reliable multi-access communication for UAV, IoT, edge devices, and mission-aware systems.
Explore research ↗Each project answers a limitation uncovered by the previous one—from a single learning client to application-, cost-, and mission-aware multi-access systems.
Introduced reinforcement learning to adapt MPQUIC path selection to wireless dynamicity.
Extended Q-learning scheduling and evaluated it in emulation and heterogeneous networks.
Combined fuzzy state representation with Q-learning for more robust decisions.
Moved from isolated devices to centralized learning across competing MPQUIC clients.
Applied maximum-entropy actor–critic learning to continuous transport measurements.
Coordinates application priorities, paths, cost, and mission requirements.
Change two wireless paths and compare policy goals. The visualization exposes the trade-off between latency, path utilization, and cellular cost.
Educational visualization—not a reproduction of paper results.
Up to 54% higher goodput and 26% shorter download time.
Up to 34.9% lower median download time than minRTT.
87.7 million targets screened across 10 ASEAN countries.
Web, video, and bulk transfer evaluated over Wi-Fi, 4G, and 5G.
Fuzzy state reasoning for heterogeneous wireless dynamicity.
7–16% lower average download time in multi-client scenarios.
Our research network connects expertise in communications, artificial intelligence, sensing, and real-world experimental systems.
You do not need prior research experience. We look for curiosity, persistence, and the willingness to build and evaluate real systems.
Build topologies, shape paths, and run reproducible tests.
Linux · PythonImplement and instrument QUIC/MPQUIC clients and servers.
Go · C/C++Train, evaluate, and explain learning-based schedulers.
Python · PyTorchCollect, clean, analyze, and visualize network-scale data.
Python · SQLTurn experiments into clear interactive dashboards.
TypeScript · UIConnect edge devices and evaluate multi-access systems.
Linux · NetworkingTell us what you are curious about, what you have built, and how much time you can commit. We will help you identify a focused first task.