Miao Pan

What they work on

1. Design software-hardware co-designed model pruning and adaptive quantization schemes that exploit the sparsity of model updates to reduce computation and memory access during federated learning on a single 5G mobile device.
2. Balance local computing against 5G transmission, and develop differentially private compression based on sparsification and quantization that protects data privacy while keeping model accuracy and communication efficiency high, evaluated on a testbed.
3. Develop a universal method to estimate the energy that diverse devices use when training deep neural networks of different architectures in federated learning, using a mobile AI testbed and energy measurement setup.
4. Improve federated learning energy efficiency in high-speed settings such as autonomous driving and augmented/virtual reality, and select training participants by weighing accuracy and latency against energy use and battery life.
5. Use serverless computing at the network edge to provide machine learning computing as a service, supporting federated learning across large numbers of heterogeneous mobile devices with bursty workloads.
6. Protect on-device training data against inference attacks with an energy-efficient piggyback differential privacy design that combines privacy gains from gradient quantization and sparsification with natural Gaussian noise in wireless channels.
7. Develop multi-bit over-the-air computation (M-AirComp) spectrum access, in which simultaneous transmissions combine in the air, to send federated learning updates efficiently under limited spectrum while reducing device energy.
8. Design collaborative training schemes between unmanned aerial vehicles (UAVs) and ground vehicles for fast, energy-efficient federated learning within the AGILE aerial-ground mobile edge computing system.
9. Jointly optimize UAV positioning, communication and computing-resource allocation, and develop dynamic UAV movement and resource-reconfiguration schemes that adapt to user demand and changing ground-vehicle resources.

More about the research

Why it matters

Federated learning lets applications in healthcare, finance and autonomous driving learn from data on people's phones and wearables without collecting that data in one place. Methods that cut energy use, network traffic and privacy risk could make these applications practical on everyday battery-powered devices without draining them quickly. They could also ease the load on crowded wireless networks with limited spectrum. Drone- and vehicle-based edge computing could bring low-delay computing to areas with many connected devices.

Known grants

From NIH and NSF research grants. Labs often have other funding we can't see.

Closest labs

The school

University of Houston: 21 labs on PI Pages, $239M a year on research, 374 research doctorates a year.

More about University of Houston

Reach out

A start for your email. Fill in the brackets.