Xin Yao

Ph.D. Candidate, Electrical and Computer Engineering

University of Florida, Gainesville, FL, USA

Google Scholar

LinkedIn

GitHub

Full CV

Email: xyao@ufl.edu

Bio: I am a Ph.D. candidate in the Department of Electrical and Computer Engineering at the University of Florida, advised by Prof. Mingyue Ji. I began my Ph.D. at the University of Utah in Aug. 2021 and moved with my advisor's research group to the University of Florida in Aug. 2024. Before that, I received my M.S. in Electrical Engineering from George Washington University (2021) and my B.Eng. in Electronic Information Engineering from Tianjin University of Science and Technology (2019). My research lies at the intersection of machine learning, wireless communication systems, and information theory. I combine theoretical analysis with algorithm and system design to develop efficient, secure, and adaptive methods for distributed learning and dynamic network control, and evaluate them on real-world wireless testbeds that I build from the physical layer up.

Short Bio

For talk introductions and conference programs.

Xin Yao is a Ph.D. candidate in Electrical and Computer Engineering at the University of Florida. His research applies machine learning to real-world wireless systems, including reinforcement learning for spectrum sharing and power allocation, multi-hop LoRa networks, wireless testbed development, decentralized learning and secure aggregation. His work focuses on learning and dynamic-control algorithms, especially on how they perform on real-world testbeds under practical constraints.

Recent News

Research Overview

My research develops learning-driven and information-theoretic methods for real-world wireless systems. I combine theoretical analysis, including Lyapunov-based optimization, convergence analysis, and information-theoretic limits, with algorithm and system design, and I evaluate the resulting methods on wireless testbeds that I build from the physical layer up. A central motivation is the gap between simulation and reality: many learning algorithms proposed for wireless systems perform well in simulation but degrade on real hardware, where carrier-frequency offsets, timing errors, interference, and hardware outages violate their design assumptions. I address this gap from both directions, with algorithms whose analysis accounts for real-world constraints and with experimental infrastructure that makes over-the-air validation systematic.

Real-World Wireless Testbeds (EdgeAI Lab)

I implement the physical-layer software stack of our group's wireless testbed in C++: modulation, demodulation, filtering, synchronization, channel encoding and decoding, and carrier-frequency and phase-offset correction. Python APIs on top of the C++ core enable scripted, repeatable over-the-air experiments, with automated workflows that adapt to heterogeneous physical-layer hardware. Our group's learning algorithms run on both lab-scale and building-scale deployments of the testbed.

UnionLabs: Shared Remote Access to NextG and IoT Testbeds

UnionLabs is an AWS-hosted platform for remote access and sharing of NextG and IoT testbeds. It includes a web interface deployed on AWS, physical-layer code for heterogeneous hardware such as USRP and LoRa, and a unified Union API that connects users' algorithms to the PHY code automatically. A single-command abstraction layer lets collaborators at different universities run their learning and optimization algorithms on multiple testbeds at the same time.

Learning-Driven Network Control

I derive optimization objectives with Lyapunov functions under real-world constraints, design state representations for reinforcement-learning control of communication systems, and deploy the resulting policies (Q-learning, DDPG, MADDPG) on live hardware. This includes spectrum sharing on a real-time LoRa testbed [C4], distributed power allocation in mmWave cellular networks [C5], and FORGE-LoRa [C1], which jointly adapts LoRa PHY parameters per slot and combines this with path discovery and optimization so that multi-hop communication stays live under interference and hardware outages. Related work includes device-free indoor localization from LoRa channel fingerprints with commodity radios [C2] and transformer-enhanced DDPG for joint beamforming and phase-shift design in RIS-assisted MU-MISO systems [C3].

Decentralized Learning and Secure Aggregation

A-DEFEAT is a decentralized learning algorithm with an adaptive error-feedback mechanism, together with the key proof technique for the adaptive mechanism and a convergence analysis that compares its rate with existing algorithms. Earlier, I studied the information-theoretic foundations of secure aggregation for federated learning when keys are shared, uncoded, among groups of users, and implemented and compared the proposed scheme against existing algorithms [C6, J1].

UAV Wireless Platform

Custom-assembled drones with integrated wireless communication payloads for airborne networking experiments: physical build, flight-controller initialization and tuning, payload design, and the communication protocol between ground devices and the airborne node.

Future Directions

Cross-layer learning-based adaptation for heterogeneous networks, with safe online learning under non-stationary conditions; digital-twin pipelines calibrated from testbed measurements, with quantitative criteria for when a policy trained in the twin can be deployed over the air; and decentralized learning over unreliable wireless networks with time-varying topologies, lossy links, and quantized communication, connected to secure aggregation under user dropouts. In parallel, I plan to develop UnionLabs into shared community infrastructure federated across institutions.

Publications

My name is underlined. See also my Google Scholar profile.

Conference Papers

Journal Articles

Submitted Manuscripts (Under Review)

Other Manuscripts

Talks

Posters & Demonstrations

Teaching & Mentoring

Honors & Awards

Professional Service