Xiao Wang (王啸)  

Phd Student

Department of Computer Science and Engineering
University at Buffalo
12 Capen Hall, Buffalo, New York 14260-1660

Email: xwang277@buffalo.edu

About Me

I am Xiao Wang, a Ph.D. student in the CSE Department at University at Buffalo, where I am advised by Dr. Venu Govindaraju. I am a part of the Center for Unified Biometrics and Sensors (CUBS) Lab, where I also work as a research assistant. My research is funded by the NSF AI Institute.

I received my M.S. degree in Computer Science from the College of Engineering and Computer Science at Syracuse University. My current research focuses on Agentic AI, with an emphasis on developing more effective and efficient ways to harness large language models as reliable, production-ready tools. In addition, I have research interests and experience in several other AI areas, including reinforcement learning, LLM post-training, and generative model, where I have also made research contributions.

Feel free to reach out to me if you are interested in collaborating. I am also actively looking for research intern positions.

News []

  • 2026.03: Two papers in AI for Education, "InterventionLens" and "ConfusionBench," were accepted to the Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR 2026).
  • 2025.6: Our paper, AutoMisty: A Multi-Agent LLM Framework for Automated Code Generation in the Misty Social Robot, has been accepted for presentation at IROS 2025.
  • 2025.05: Started my summer research intern journey at Microsoft MSAI PhD Research Interns at Reasoning and LU Group.
  • 2024.09: Our paper, ProxyFusion: FaceFeature Aggregation Through Sparse Experts, has been accepted for presentation at NeurIPS 2024.
  • 2024.09: Our paper on 3D sign language motion generation has been accepted by EMNLP 2024.
  • 2024.08: A paper on enhancing facial expression inference has been accepted by the ECCV 2024 Workshop.
  • 2024.08: Started my Ph.D. journey at University at Buffalo, SUNY,Buffalo, NY.
  • 2024.03: Our paper "SignAvatar" has been accepted by FG 2024 .
  • 2024.03: I have been admitted to the Fall 2024 Doctorate of Philosophy program in CSE at the University at Buffalo, with a Research Assistant position at the Center for Unified Biometrics and Sensors under the supervision of Professor Venu Govindaraju.
  • 2023.06:Research Assistant at University at Buffalo's Human Behavior Modeling lab, under the direct supervision of Professor Ifeoma Nwogu, focused on implementing AI generative algorithms for studying human behaviors.
  • 2023.05: Successfully completed my Master's degree at Syracuse University with 3.861 / 4 GPA, Syracuse, NY.
  • 2021.08: Started my Master's journey at Syracuse University, Syracuse, NY.

Selected Research

AutoMisty: A Multi-Agent LLM Framework for Automated Code Generation in the Misty Social Robot
 Xiao Wang*, Lu Dong*, Sahana Rangasrinivasan, Ifeoma Nwogu, Srirangaraj Setlur, Venugopal Govindaraju
International Conference on Intelligent Robots and Systems (IROS 2025)
DOI PDF BibTeX Project Webpage
InterventionLens: A Multi-Agent Framework for Detecting ASD Intervention Strategies in Parent-Child Shared Reading
 Xiao Wang*, Lu Dong*, Ifeoma Nwogu, Srirangaraj Setlur, Venu Govindaraju
Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR 2026)
arXiv PDF BibTeX Project Webpage
ConfusionBench: An Expert-Validated Benchmark for Confusion Recognition and Localization in Educational Videos
 Xiao Wang*, Lu Dong*, Mark G Frank, Srirangaraj Setlur, Venu Govindaraju, Ifeoma Nwogu
Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR 2026)
DOI PDF BibTeX Project Webpage
Ig3D: Integrating 3D Face Representations in Facial Expression Inference
 Xiao Wang*, Lu Dong*, Srirangaraj Setlur, Venu Govindaraju, Ifeoma Nwogu
Proceedings of the 18th European Conference on Computer Vision (ECCV 2024)
DOI PDF BibTeX Project Webpage
Word-Conditioned 3D American Sign Language Motion Generation
 Lu Dong, Xiao Wang, Ifeoma Nwogu
The 2024 Conference on Empirical Methods in Natural Language Processing (EMNLP 2024)
DOI PDF BibTeX Project Webpage
ProxyFusion: Face Feature Aggregation Through Sparse Experts
 Bhavin Jawade, Alexander Stone, Deen Dayal Mohan, Xiao Wang, Srirangaraj Setlur, Venu Govindaraju
The Thirty-eighth Annual Conference on Neural Information Processing Systems (NeurIPS 2024)
DOI PDF BibTeX Project Webpage
SignAvatar: Sign Language 3D Motion Reconstruction and Generation
 Lu Dong, Lipisha Chaudhary, Fei Xu, Xiao Wang, Mason Lary, Ifeoma Nwogu
The 18th IEEE International Conference on Automatic Face and Gesture Recognition (FG 2024)
DOI PDF BibTeX Project Webpage

Experience

Academic Services

Journal Reviewer

Highlighted Projects

Revolutionizing Therapy with Robot Agents
This project unifies three core capabilities—(1) the planning, reasoning, and automatic code-generation power of large language models; (2) the environmental understanding afforded by vision–language models; and (3) real-time intent recognition from speech and pose—and deploys them on the Misty social robot. Guided by the MistyCoPilot scheduler, which automatically invokes Automisty-generated code, the system can flexibly compose functional modules on demand, enabling highly abstract and complex human–robot interaction tasks in speech-language pathology (SLP) scenarios.

Please check the title link for more details.
ARL-INTERSECT: A Multi-Agent Reinforcement Learning Algorithm for Autonomous Cars
This work introduces ARL-INTERSECT, a multi-agent reinforcement learning algorithm designed for autonomous vehicles navigating unsignalized intersections. By training independent Actor-Critic models for each vehicle agent and leveraging parallel computing for faster data sampling, we improve training stability and scalability. Integrating A3C with PPO and incorporating entropy-based exploration yields smoother convergence and more reliable policy enhancements. Evaluated in a simulated intersection scenario, ARL-INTERSECT outperforms traditional independent A3C approaches in terms of convergence speed, policy performance, and interaction safety. Moreover, it demonstrates adaptability and scalability across both small-scale (4 agents) and large-scale (10 agents) environments.
Please follow the title link for access to my Project codes and Demo.
Predicting One-step Proofs in Formal Mathematics using Transformer-based Neural Networks
This research explores the use of transformer models, like OpenAI's ChatGPT, for Automated Theorem Proving (ATP) on the Metamath dataset (Set.mm). By framing proof verification as a sequence-to-sequence task, we leverage the logical relationships in mathematical proofs. Our approach highlights AI's potential in theorem proving, addressing key challenges and demonstrating promising results in complex mathematical tasks.

Please follow the title link for access to the source code and detailed reports.
XiaoStyle - Customizable & Secure eCommerce Platform
This project is a custom eCommerce application built with Python Django. It features a custom user model, product and category management, cart operations, an unlimited product image gallery, order processing, and payment integration. Post-order functionalities include stock reduction, email notifications, and invoice generation. The site also offers a review and rating system with interactive stars, account management, session handling, and secure user authentication with token-based verification and password resets.

Please follow the title link for access to the source code and YouTube Demo.
Ultimate Data Navigator: Efficient Data Management System
Leveraging C++ for high-performance computing, The Ultimate Data Navigator automates scanning, data gathering, and uploads to a centralized database, while its modular utilities manage scheduling, resource allocation, file transfer, and database operations across diverse data types and volumes. Built on a custom developer framework, the system allows rapid deployment and easy modifications. Additionally, the solution is cost-effective, efficiently processing millions of data entries weekly using free, open-source components.

Please follow the title link for access to the source code and detailed reports.
Optimizing Factory Assembly Line Through C++ Multi-threading
This project utilizes C++ multi-threading, including thread conditional variables and atomic operations, to simulate and optimize factory assembly line strategies. After identifying inefficiencies, a new approach was implemented using Object-Oriented Design (OOD) with an analytic module added for result analysis.

Please follow the title link for access to the source code and detailed report.
Robust Remote Process Control system
This project develops a remote process control system utilizing threading, network communication via UDP, signal handling, and process management. The core component is responsible for overseeing processes based on a configurable setup. It efficiently handles process termination requests and allows for remote control of process shutdown.

Please follow the title link for access to the source code and detailed reports.
Implementing and Securing Network Systems
This lab series, based on the SEEDLAB project by Professor Du Wenliang, involves implementing and analyzing network protocols like ARP, IP, UDP, TCP, and more. The work includes VPN and firewall operations, DNS and BGP studies, and security assessments of potential vulnerabilities.

Please follow the title link for access to the source code and detailed reports.
RecruitPro: HR Recruitment Database Management System
RecruitPro is a powerful HR database that optimizes candidate tracking throughout the entire hiring process. The system efficiently manages job postings, candidate profiles, interviews, evaluations, reimbursements, and onboarding, while adhering to 3rd Normal Form principles to ensure data integrity and security. It leverages advanced database techniques such as views, stored procedures, functions, triggers, and transactions, alongside user role management, to provide a comprehensive and secure solution for HR operations.

Please follow the title link for access to the source code and detailed reports.

Last Updated on April 7, 2026