About

Heng Xu is a faculty member associated with the Center for Human-Computer Interaction at Penn State. The Center is an interdisciplinary organizational unit dedicated to research, instruction, and outreach in human-computer interaction (HCI). It aims to facilitate interdisciplinary faculty collaboration across various departments, focusing on a broad range of HCI issues, problems, and opportunities. The Center's research encompasses areas such as software and information design, end-user programming, design rationale, creativity in design, educational technology, collaborative learning, open source software, e-science, web-based systems, online communities, wireless networks, decision support, geospatial information systems, usability engineering, and theories in HCI. The faculty involved in the Center conduct both basic and applied research, contributing to design methods, multimedia access, digital libraries, community computing, visualization, virtual environments, input/output ergonomics, and usability evaluation. Heng Xu's work supports the development of innovative educational programs, promotes interdisciplinary research, and enhances the university's prominence in HCI, with active engagement in outreach and collaboration with industry and the broader community.

Research topics

• Computer Science
• Psychology
• Artificial Intelligence
• Political Science
• Machine Learning
• Computer Security
• Medicine
• Clinical psychology
• Law
• Mathematics
• Social psychology
• Public relations
• Business
• Knowledge management
• Epistemology
• Statistics

Selected publications

• A Novel Scheme for Recommendation Unlearning Verification (RUV) Using Non-Influential Trigger Data

  2025-01-10

  article

  Machine unlearning has garnered widespread attention, due to various reasons, including privacy-preserving, model usability, and legal regulations. It requires model providers to unlearning users' data from models upon receiving unlearning request. Recommendation systems have also been extensively researched in the field of deep learning, particularly within the context of big data environments. However, little research can be found to verify the effectiveness of unlearning approach using pure tabular data-based recommendation scenario. In this paper, we propose a recommendation unlearning verification (RUV) scheme based on non-influential trigger data, which fills this gap. Users can use the recommendation rate for selected target items to determine whether the recommendation system complies with unlearning requests. Evaluation results on real datasets confirm the efficiency and effectiveness of our proposed RUV scheme.

  PublisherDOI

• KONTROL: Offloading Data-Driven Congestion Control Intelligence for IoT Networks

  IEEE Internet of Things Journal · 2025-10-30

  article

  Congestion control is a cornerstone of reliable data transport, but current fixed-rule algorithms struggle with the diverse link characteristics of Internet of Things deployments. A more critical challenge is that many resource-constrained Internet of Things devices lack the computational power to run advanced, data-driven methods locally, hindering performance and adaptability. To address this, we introduce KONTROL, a service framework that decouples congestion control intelligence from end devices by offloading the decision-making logic to a centralized congestion control server. This enables lightweight clients to leverage sophisticated control strategies without bearing the computational burden. As a critical instance within our framework, we implement Deep Reinforcement Learning agents on the server, which learn to optimize window adjustments for each sender based on real-time relayed network statistics via kernel modifications. Our experimental evaluation across challenging simulated environments shows that the proposed scheme achieves consistently high throughput and low latency while maintaining fairness. These results validate the viability of the offloading paradigm, paving the way for more flexible transport protocols for the broader Internet of Things ecosystem.

  PublisherDOI

• Testing of Reverse Causality Using Semi-Supervised Machine Learning

  Psychometrika · 2025-04-07

  articleOpen access

  Abstract Two potential obstacles stand between the observation of a statistical correlation and the design (and deployment) of an effective intervention, omitted variable bias and reverse causality . Whereas the former has received ample attention, comparably scant focus has been devoted to the latter in the methodological literature. Many existing methods for reverse causality testing commence by postulating a structural model that may suffer from widely recognized issues such as the difficulty of properly setting temporal lags, which are critical to model validity. In this article, we draw upon advances in machine learning, specifically the recently established link between causal direction and the effectiveness of semi-supervised learning algorithms, to develop a novel method for reverse causality testing that circumvents many of the assumptions required by traditional methods. Mathematical analysis and simulation studies were carried out to demonstrate the effectiveness of our method. We also performed tests over a real-world dataset to show how our method may be used to identify causal relationships in practice.

  PublisherOA PDFDOI

• Study on flow field and pressure response characteristics of ships of different scales under periodic waves

  SSRN Electronic Journal · 2025-01-01

  preprintOpen access
  PublisherDOI

• Let the Time Speak: Merger by Digital Platforms and Dynamic Review

  SSRN Electronic Journal · 2025-01-01

  preprintOpen access1st authorCorresponding
  PublisherDOI

• Temporal-Aware GPU Resource Allocation for Distributed LLM Inference via Reinforcement Learning

  ArXiv.org · 2025-07-14

  preprintOpen access

  The rapid growth of large language model (LLM) services imposes increasing demands on distributed GPU inference infrastructure. Most existing scheduling systems follow a reactive paradigm, relying solely on the current system state to make decisions, without considering how task demand and resource availability evolve over time. This lack of temporal awareness in reactive approaches leads to inefficient GPU utilization, high task migration overhead, and poor system responsiveness under dynamic workloads. In this work, we identify the fundamental limitations of these instantaneous-state-only scheduling approaches and propose Temporal Optimal Resource scheduling via Two-layer Architecture (TORTA). TORTA introduces a spatiotemporal scheduling framework that captures both long-term workload patterns and short-term execution constraints. It adopts a two-layer design: a macro-level scheduler leverages reinforcement learning and optimal transport to coordinate inter-region task distribution, while a micro-level allocator refines task-to-server assignments within each region to reduce latency and switching costs. Experimental results across multiple network topologies show that TORTA reduces average inference response time by up to 15\%, improves load balance by approximately 4-5\%, and cuts total operational cost by 10-20\% compared to state-of-the-art baseline methods.

  PublisherOA PDFDOI

• The Personalized Scene-Based AI Agents: Exploring Chinese Language Learning Among International Students

  2025-05-14

  article

  With the development of deep learning and large language model (LLM) technologies, AI agents, as a key application of generative artificial intelligence (AIGC), have demonstrated significant potential in the field of international Chinese education. Personalized scenario AI agents, leveraging open-source AI chat frameworks, large-scale Chinese language corpora, and dynamic adjustment mechanisms (RAG, PAL, ReAct), effectively address key challenges in Chinese language acquisition for international students. These challenges include a lack of communicative practice, difficulties in acquiring everyday expressions, the inability of large-class instruction to accommodate individual differences, and the absence of real-time feedback. By constructing immersive interactive scenarios, personalized learning pathways, and intelligent feedback systems, AI agents enhance learning efficiency and flexibility while providing new pathways for the intelligent development of international Chinese education.

  PublisherDOI

• Asset Pricing and Portfolio Investment Management Using Machine Learning: Research Trend Analysis Using Scientometrics

  Economics · 2024-01-01 · 3 citations

  articleOpen access

  Abstract “Asset pricing” in the context of financial economics pertains to the investigation and formulation of two fundamental pricing ideas and the models that go along with them. Various models exist for different scenarios, but they can be traced back to either general equilibrium asset pricing or rational asset pricing. Asset pricing models, as the name suggests, serve as valuable tools to assess the value of assets. The general equilibrium theory states that supply and demand interact to determine market prices. In this context, asset prices collectively satisfy the market clearing condition, which dictates that the supply and demand for each asset are equal at the prevailing price. Another crucial aspect of financial planning is portfolio management (PM), which aims to maximise investment profits while minimising losses. PM involves implementing effective asset allocation strategies to enhance returns and mitigate risks. Numerous studies have been conducted worldwide on various types of asset pricing models and investment portfolios, with some incorporating machine learning and deep learning techniques. In several models, the predictive accuracy has exceeded 90%. To shed light on the current research landscape in the realm of asset pricing and portfolio investment, we conducted a scientometric analysis.

  PublisherOA PDFDOI

• The impact of environmental, social, and governance on corporate financial performance: Empirical evidence based on Chinese a-share listed companies

  International Journal of Applied Economics Finance and Accounting · 2024-01-12 · 5 citations

  articleOpen access

  This study aims to delve into the influence of ESG performance on the financial outcomes of companies listed on China's A-share market, emphasizing the interplay of ESG's three critical dimensions: environmental, social, and governance performance. Utilizing ESG data from A-share listed companies in China spanning from 2013 to 2022, regression analysis was executed in STATA 17.0. Factors like company size, leverage, growth, age, board size, and ownership concentration were integrated as control variables. The results underscored a positive association between both holistic ESG performance and its individual dimensions (environmental, social, and governance performance) and financial outcomes. Notably, non-state-owned enterprises exhibited a more pronounced positive relationship between ESG performance and financial results than their state-owned counterparts. Drawing from these insights, it's advocated that companies amplify their efforts towards ESG performance enhancement. It further accentuates the need for regulatory bodies to formulate pertinent policies and amplify oversight. Additionally, investors are advised to incorporate ESG performance metrics into their investment decisions, promising not only improved financial standing for corporations but also fostering sustainability and comprehensive growth in the social, environmental, and economic domains.

  PublisherOA PDFDOI

• Recommendation System Model Ownership Verification via Non-Influential Watermarking

  2024-12-02

  article

  While deep learning-based recommendation systems have achieved great success, recommendation system models are also at serious risk of intellectual property infringement. Current model watermarking research faces significant challenges in terms of fidelity, invisibility, and efficiency. Additionally, existing model watermarking techniques are predominantly applied to image data, with limited applicability to tabular data. In this paper, we introduce an innovative watermarking framework designed to safeguard the ownership of recommendation system models. Specifically, we verify recommendation system model ownership by embedding a type of backdoor watermark into the training dataset, which does not affect model performance. We have conducted experiments on several classical datasets to validate the reliability and effectiveness of our approach.

  PublisherDOI

Recent grants

• CT-ER: Privacy Assurance in Location-Based Services: Integrating Economic Exchange and Social Justice Perspectives

  NSF · $175k · 2007–2010

• RR: Establishing and Boosting Confidence Levels for Empirical Research Using Twitter Data

  NSF · $400k · 2018–2018

Frequent coauthors

• Mary Beth Rosson

  Pennsylvania State University

  23 shared

• John M. Carroll

  19 shared

• Pamela Wiśniewski

  18 shared

• Hock‐Hai Teo

  SEGi University

  12 shared

• Tamara Dinev

  Florida Atlantic University

  10 shared

• Pan Shi

  Southwest Medical University

  10 shared

• Rachida Parks

  9 shared

• Yilu Zhou

  Fordham University

  9 shared

Labs

• HCI CenterPI

  Human-Computer Interaction research