About

Kun Qian is an Assistant Professor in the Department of Computer Science at the University of Virginia. His research interests lie in wireless and mobile technologies that enable ambient intelligence and ubiquitous connectivity. He has designed wireless models, algorithms, and systems with applications in next-generation communication, smart IoT sensing, autonomous driving, and mobile computing.

Research topics

• Artificial Intelligence
• Computer Science
• Geography
• Remote sensing
• Machine Learning
• Computer vision
• Real-time computing
• Telecommunications
• Speech recognition
• Mathematics
• Meteorology
• Environmental science

Selected publications

• BeamFormer Artifact — MobiSys 2026

  Zenodo (CERN European Organization for Nuclear Research) · 2026-04-12

  otherOpen access

  This artifact contains the source code, pre-trained model weights, and datasets for reproducing the experimental results in "BeamFormer: Transformer-based Beam Management for 6G Networks" (MobiSys 2026). This code covers the version submitted for Artifact Evaluation. Additional experiments have since been added to the paper; for the AE code covering all experiments in the latest version of the paper, please refer to: https://github.com/Shunqiang-Feng/BeamFormer/tree/AE

  PublisherDOI

• BeamFormer Artifact — MobiSys 2026

  Zenodo (CERN European Organization for Nuclear Research) · 2026-04-12

  otherOpen access

  This artifact contains the source code, pre-trained model weights, and datasets for reproducing the experimental results in "BeamFormer: Transformer-based Beam Management for 6G Networks" (MobiSys 2026). This code covers the version submitted for Artifact Evaluation. Additional experiments have since been added to the paper; for the AE code covering all experiments in the latest version of the paper, please refer to: https://github.com/Shunqiang-Feng/BeamFormer/tree/AE

  PublisherDOI

• Automated Compliance Monitoring: A Machine Learning Approach for Digital Services Act Adherence in Multi‐Product Platforms

  Preprints.org · 2025-04-28

  preprintOpen accessSenior author

  This paper presents an innovative machine learning approach for automated compliance monitoring of Digital Services Act (DSA) requirements across multi-product digital platforms. The proposed framework addresses the significant challenges of monitoring regulatory compliance in complex digital environments where manual verification processes prove insufficient and error-prone. The methodology introduces a formalized representation of DSA requirements through algorithmic processing and transforms these into machine-verifiable specifications using metamorphic testing principles and timed automata models. The core architecture implements a hybrid risk assessment model combining supervised and unsupervised learning techniques to evaluate compliance across heterogeneous platform environments. Comprehensive evaluation across multiple digital service categories demonstrates detection accuracy between 0.86-0.94 (F1-score) with processing efficiency ranging from 78% to 95% depending on platform characteristics. The multi-platform data integration pipeline achieves near real-time monitoring capabilities while respecting data protection constraints. The framework addresses key technical challenges including the complexity of requirement formalization, data access limitations, and adaptation to evolving regulatory interpretations. This research contributes significant advancements toward automated, scalable compliance verification solutions essential for effective implementation of the Digital Services Act across diverse digital service ecosystems.

  PublisherOA PDFDOI

• Toward Spoofing-Resilient and Communication-Integrated MmWave Radar Sensing

  2025-06-23

  articleOpen access1st authorCorresponding

  MmWave FMCW radars are integrated into many sensing systems for robust sensing. However, their sensing functions are vulnerable to spoofing attacks and interfered with by backscatter communications, both of which can cause sensor malfunction and system failure. Noticing that radar spoofing and communication share similar signal modulation mechanisms, in this paper, we present SCR, a new Spoofing-resilient and Communication-integrated Radar sensing scheme. SCR is based on the rigorous analysis of the radar sensing model that highlights the differences between modulated spoofing and communication signals and normal sensing signals reflected by natural objects. The key designs of SCR are a novel chirp configuration scheme and signal processing pipeline, which signify different patterns between modulated and normal signals in radar spectra, for reliable detection of spoofing and communication. We have developed SCR and tested it with actual 77 GHz mmWave radar sensors and backscatter prototypes. Our field tests show that SCR can reliably detect fake objects created by modulated signals in both velocity and distance radar sensing domains.

  PublisherOA PDFDOI

• NeuroRadar: A Neuromorphic Radar Sensor for Low-Power IoT Systems

  Communications of the ACM · 2025-08-21

  articleOpen access

  A Neuromorphic Radar Sensor for Low-Power IoT SystemsRadar sensors in Internet of Things (IoT) systems have gained traction in recent years and are widely used in healthcare, smart homes, industrial automation, and intelligent transportation.The high power consumption of radar hardware remains a significant challenge, particularly for battery-operated IoT devices and wearables, where energy efficiency and battery lifespan are crucial.Compounding this issue, numerous smart-sensing applications-such as motion-activated security radar, wearable gesture recognition, and activity classification-often employ power-intensive artificial neural networks (ANNs) for signal processing.Unlike human neurons that operate in short, pulse-based bursts, ANNs prolong the activity of their "neurons" using continuous activation functions, which substantially increases the power demands of IoT devices.Furthermore, ANNs use the classical von Neumann architecture, which frequently shuttles data between physically separate CPU and memory units, resulting in additional processing overhead.Recent advances in neuromorphic engineering have inspired spiking neural networks (SNNs) and dedicated neuromorphic circuits 8 that better approach the efficiency of sensory signal processing in the brain.SNNs are structured to mirror the pulse-based behavior of the human nervous system.They consist of spiking neurons and the synaptic connections between them.Realized on dedicated neuromorphic circuits, SNNs showcase exceptional energy efficiency that surpasses traditional von Neumann computing units by orders of magnitude. 4The revolution in neuromorphic computing has also given rise to state-of-the-art neuromorphic sensing hardware, such as the energy-efficient, fast-response event camera. 11nspired by these advances, recent research has proposed SNN-based signal processing to facilitate low-power radar operation. 2,3However, these systems do not incorporate a full-fledged neuromorphic hardware architecture.Primarily, the analog front end of these SNN radar systems 2,3 remains the same as traditional radars.Although SNN-based signal processing has lowered signal-processing power consumption to the order of hundreds of W 2 , the radar front end can demand tens to hundreds of milliwatts.This discrepancy poses a challenge to achieving truly energy-efficient radar sensing.Additionally, SNN radar systems 2,3 continue to rely on conventional CPUs or digital signal processing (DSP) units for signal processing.The radar signals must be first sampled by analog-to-digital converters (ADCs), mapped into spikes, and then processed by SNNs for ranging or environmental perception.Unfortunately, the extra sampling steps prior to the SNN involve traditional computing units, which adds a substantial overhead, underutilizing neuromorphic computing's full potential.In this paper, we introduce NeuroRadar, a novel low-power radar-sensing system that fully exploits the power of neuromorphic sensing and computing.NeuroRadar draws inspiration from neuromorphic sensors that mimic mammalian sensory systems, generating event-triggered

  PublisherOA PDFDOI

• Radio Frequency Ray Tracing with Neural Object Representation for Enhanced RF Modeling

  2025-06-10 · 2 citations

  article

  Radio frequency (RF) propagation modeling poses unique electromagnetic simulation challenges. While recent neural representations have shown success in visible spectrum rendering, the fundamentally different scales and physics of RF signals require novel modeling paradigms. In this paper, we introduce RFScape, a novel framework that bridges the gap between neural scene representation and RF propagation modeling. Our key insight is that complex RF-object interactions can be captured through object-centric neural representations while preserving the composability of traditional ray tracing. Unlike previous approaches that either rely on crude geometric approximations or require dense spatial sampling of entire scenes, RFScape learns perobject electromagnetic properties and enables flexible scene composition. Through extensive evaluation on real-world RF testbeds, we demonstrate that our approach achieves 13 dB improvement over conventional ray tracing and 5 dB over state-of-the-art neural baselines in modeling accuracy, while requiring only sparse training samples.

  PublisherDOI

• Shop-R1: Rewarding LLMs to Simulate Human Behavior in Online Shopping via Reinforcement Learning

  arXiv (Cornell University) · 2025-07-23

  preprintOpen access

  Large Language Models (LLMs) have recently demonstrated strong potential in generating 'believable human-like' behavior in web environments. Prior work has explored augmenting training data with LLM-synthesized rationales and applying supervised fine-tuning (SFT) to enhance reasoning ability, which in turn can improve downstream action prediction. However, the performance of such approaches remains inherently bounded by the reasoning capabilities of the model used to generate the rationales. In this paper, we introduce Shop-R1, a novel reinforcement learning (RL) framework aimed at enhancing the reasoning ability of LLMs for simulation of real human behavior in online shopping environments. Specifically, Shop-R1 decomposes the human behavior simulation task into two stages: rationale generation and action prediction, each guided by distinct reward signals. For rationale generation, we leverage internal model signals (e.g., logit distributions) to guide the reasoning process in a self-supervised manner. For action prediction, we propose a hierarchical reward structure with difficulty-aware scaling to prevent reward hacking and enable fine-grained reward assignment. This design evaluates both high-level action types and the correctness of fine-grained sub-action details (attributes and values), rewarding outputs proportionally to their difficulty. Experimental results show that our method achieves a relative improvement of over 65% compared to the baseline. The project page is available at https://damon-demon.github.io/shop-r1.html.

  PublisherOA PDFDOI

• Combined Vs. Single Supplementation of L-Citrulline and Sodium Bicarbonate During Sprint Interval Training in Basketball Players: Effects on Physical Performance and Hormonal Adaptations

  Journal of Sports Science and Medicine · 2025-03-24 · 2 citations

  articleOpen accessSenior author

  This study examined the effects of L-citrulline (L-CIT) and sodium bicarbonate (Sb) supplementation during short sprint interval training (SSIT), both individually and combined, over a 28-day period, to assess the impact on the physical performance and hormonal adaptations of basketball players. Forty young (age = 21.5 ± 1.7 years) male basketball players were randomly assigned into five groups of eight subjects including either L-CIT (6 g daily), Sb (0.3 g·kg-1 of Sb daily), L-CIT+Sb, placebo (PL), and or an active control group (CON). The training groups engaged in SSIT sessions, consisting of three sets of ten repetitions of five-second all-out sprints, three times per week over four weeks. A series of physical performance tests including countermovement vertical jump, a 20-m linear sprint, Illinois change of direction speed, Wingate anaerobic power, and an incremental exercise test were conducted before and after the training period. In addition, blood samples were obtained to analyze resting testosterone and cortisol levels before and after the training period. Significant improvements in physical performance were observed across all training groups after the 4-week intervention (p < 0.05). Notably, the groups receiving supplements exhibited more pronounced improvements in the physical performance tests (p < 0.01) in comparison to the PL group, indicating superior adaptations. In addition, no significant (p > 0.05) changes were seen in testosterone, but the supplement groups showed lower cortisol levels than other groups following the 4-week intervention. There were no significant differences in physical performance adaptations among the supplement groups. The study reveals that short-term supplementation of L-citrulline and sodium bicarbonate in the capsule form effectively enhance physical performance of basketball players in comparison to a placebo.

  PublisherDOI

• Development of Badminton Tactical Simulation and Scenario Training System Based on Generative AI

  2025-05-16

  article

  At present, badminton tactical training has problems such as single confrontation scenario and insufficient dynamic strategy generation. Traditional simulation methods rely too much on coaching experience and fixed playback videos, which is difficult to meet the needs of athletes’ personalized tactical awareness training and adaptive training in complex confrontation environments. In response to the above problems, this study proposes a tactical simulation and scenario training system based on generative AI. First, this paper constructs a multidimensional tactical knowledge graph and establishes a tactical ontology model including spatial trajectory, hitting mode, and confrontation strategy; secondly, a dynamic confrontation scenario generation module based on conditional generative adversarial network (CGAN) is developed. Finally, a tactical optimization engine driven by the Proximal Policy Optimization (PPO) algorithm is designed to dynamically adjust the difficulty coefficient and strategy complexity of the generated scenarios. In the experiment, the athletes in the experimental group using the system achieved an accuracy rate of 85.4% in tactical prediction, which is significantly higher than the 75.1% of the control group; in terms of multi-shot connection error rate, the experimental group is 12.0 %,while the control group is17.2 %. These results show that this system can provide a personalized and diversified training environment and significantly improve athletes’ tactical awareness and technical connection ability.

  PublisherDOI

• The whistle‐blower effect vs. the cry‐wolf effect: A game analysis framework for collaborative epidemic information governance

  Risk Analysis · 2025-01-17 · 1 citations

  articleOpen access

  The unpredictability of the epidemics caused by new, unknown viruses, combined with differing responsibilities among government departments, often leads to a prisoner's dilemma in epidemic information governance. In this context, the whistle-blower effect in the health departments leads to delayed reporting to avoid potential retaliation, and the cry-wolf effect in the administrative departments results in sustained observation to avoid ineffective warnings. To address these challenges, we employ game theory to analyze the dynamics of epidemic information governance and focus on two external governance mechanisms-superior accountability and media supervision-that can help resolve the prisoner's dilemma during and after an outbreak. Our analysis indicates that it is necessary to increase the strategic coordination of whistle-blowers in the short-term decision-making during the outbreak. From a long-term evolution perspective, maintaining optimal levels of superior accountability and media supervision is essential to overcoming the prisoner's dilemma. Media supervision works more slowly in the implement effectiveness than more direct superior accountability. This paper highlights the crucial roles of the whistle-blower effect and the cry-wolf effect in coordination failures of epidemic information governance during outbreaks of unknown viruses. It clarifies the strategic coordination pathways between expert systems and bureaucratic systems and emphasizes the importance of superior accountability and media supervision to enable effective, collaborative epidemic information governance.

  PublisherOA PDFDOI

Frequent coauthors

• Zheng Yang

  Tsinghua University

  49 shared

• Chenshu Wu

  47 shared

• Yi Zhang

  33 shared

• Yunhao Liu

  15 shared

• Qi Yu

  13 shared

• Yanchen Liu

  Hunan University

  11 shared

• Xinyu Zhang

  11 shared

• Rui Guo

  11 shared

Education

• Ph.D., Software Engineering

  Tsinghua University

  2019

• B.S., Software Engineering

  Tsinghua University

  2014

Awards & honors

• SenSys Best Paper Award 2023
• MobiCom Best Paper Award 2020