Research Projects

  • AIoT Optimization: One of the fundamental challenge of the Cyber Physical Systems is throughput. In order to address the fundamental challenge of proximal wireless interference, Dr. Xu designed a three-step optimization framework, namely topology, routing, and packet-level scheduling, and considered the three correlated topics: a) reducing  throughput to maximum weight independent set (MWIS); b) maximize packet-level transmissions via coverage, packing, and connectivity; c) emerging techniques such as 5G, MCMR (multi-channel multi-interface), MIMO (multi-channel input and output), Duty Cycle to improve throughput further. Regarding topic a), Dr. Xu introduces a zero-sum game framework to establish the reduction between the shortest scheduling to MWIS. Dr. Xu proposes a new paradigm to replace the traditional multi-hop transmission algorithms in the past few decades, and it also provides theoretical analysis tools for many other network flow problems. Regarding topic b), Dr. Xu addresses the limitations of traditional algorithms and proposes a new algorithm under the physical interference model. Most of the previous work relied on a simplified version of the protocol interference model, and the work of Dr. Xu is more realistic and challenging. Regarding topic c), Dr. Xu explores the connections betweennew emerging techniques and traditional scheduling and proposes approximate algorithms in this field.

  • Reinforcement Learning for IoT: Spectrum scarcity and low utilization is a grand challenge. Dynamic spectrum access can effectively resolve this contradiction. Dr. Xu considered three correlated topics: a) online algorithms with no info., b) online learning and algorithms under probability distribution, c) Online learning and algorithms under dynamic changes. Regarding topic a), Dr. Xu proposed a preemption and compensation plan, designed an algorithm based on the compensation function, and proved that the performance was approximately optimal. Regarding topic b), Dr. Xu proposes a series of algorithms considering most previous studies did not consider interference constraints. Regarding topic c), Dr. Xu started from the framework of restless multi-arm bandit (MAB) and stochastic optimization, from unknown to partially visible, defined the expected information state, and designed an approximate algorithm under interference constraints for the first time. MAB has broad application in unmanned multi-body system tracking, network security, 5G and other fields.

  • Game Theory and Economics: Considering the tradeoff between selfish users and global welfare, how to design the pricing and allocation of spectrum resources to encourage users to bid truthfully is the key. Dr. Xu has studied three topics: a) The first-price auction game, that is, each rational user bids based on his evaluation, and the charge to the winning user is equal to his bid; b) The second-price auction game, that is, the price of the winning user is equal to the second-highest bid; c) Stackelberg model game. Regarding topic a), Dr. Xu applies a "smooth analysis" to the game theory and greatly reduce complexity and obtains the first approximate Nash equilibrium under interference constraints. Regarding topic b), Dr. Xu uses a pricing mechanism to incentivate truthful bidding by participating users. Regarding topic c), Dr. Xu design strategies of the two groups of game users.

  • With Bell Labs:I collaborated with security group on a VoIP project. I worked on exploring the Content Addressable Network (CAN) Distributed Hash Table (DHT). One goal is to create an empirical implementation and measure the performance of an experiment CAN network to compare it to the classical results. Our results compared well to the paper by Ratnasamy et al., thus providing one of the first independent empirical validation of a CAN DHT. We also implemented a visualization component that demonstrated the CAN during its formation and when nodes leave the CAN (both scheduled departures and unscheduled departures). The work we started has been used as a base for other projects at Bell Labs.

  • With REAI Inc: I collaborate with REAI Inc. to provide the proprietary AI platform for the next generation real estate industry. We have an uique team of top ML/AI experts with Georgia Tech PhD, as well as Harvard MBA and Wharton MBAs, and have put together the cutting edge AI algoritm to solve the fundermental problems in real estate indstry, starting from the well known "Remorse Factor" that buyers regret after moving in realizing the property is not the best for their needs. Our REAI platform provides unique AI Smart Matching solution, that address these industry problems including buyer having difficulty locating the most appropriate property, agents having difficulty finding appropriate quality leads/clients, etc.

    Recent Grants
  • 课题负责人,安徽省科技重大专项项目,自动驾驶系统网络安全关键技术研究与应用。金额,400万,批准号:S202103a05020098, 开始时间:2022年1月-2025年12月。
  • 主持,面上项目,云边端协同的智能物联网在线资源调度与性能优化。金额:58万,批准号:62172383, 开始时间:2022年1月-2025年12月。
  • PI, Research Grant, SunTrust Banks, Feb 13, 2020 (Amount: $9, 000.00).
  • Co-PI, ALG Grant, State of Georgia, Jan 31, 2020 (Amount: $30, 000.00).
  • PI, NSF ECCS-1523965, Optimal Joint Spectrum Allocation and Scheduling for Cognitive Radio Networks, (Amount: $244, 808.00).