About me

I am a first-year PhD student in the Manning College of Information & Computer Sciences at UMass Amherst, advised by Professor Mingda Qiao. I received my B.S. in Computer Science from Rensselaer Polytechnic Institute, where I had the privilege of participating in various research programs supervised by Professor Lei Yu, Bülent Yener, and Jianxi Gao.

Research

I like both theory and application research. My recent projects are:

• Learning Theory: Theoretical aspects of prediction, learning, and decision-making in online settings.
• Trustworthy ML: Adversarial Robustness of neural networks against poison and backdoor attacks.
• AI for Science: Biosignal/EEG analysis; LLM-based multi-agent for healthcare.

Publication

• On the adversarial robustness of graph neural networks with graph reduction [pdf] [code]

  Kerui Wu, Ka-Ho Chow, Wenqi Wei, Lei Yu

  Accepted by ESORICS 2025 (Acceptance rate: 16%)

• On the Robustness of Graph Reduction Against GNN Backdoor [pdf]

  Yuxuan Zhu, Michael Mandulak, Kerui Wu, George Slota, Yuseok Jeon, Ka-Ho Chow, Lei Yu

  Accepted by AISec’24 CCS 2024 (Acceptance rate: 25%)

• Large EEG-U-Transformer for Time-Step Level Detection Without Pre-Training [pdf] [code]

  Kerui Wu, Ziyue Zhao, Bülent Yener

  Preprint on arXiv

• ClinicalReTrial: A Self-Evolving AI Agent for Clinical Trial Protocol Optimization [pdf]

  Sixue Xing, Xuanye Xia, Kerui Wu, Meng Jiang, Jintai Chen, Tianfan Fu

  Preprint on arXiv

Awards

• 1st Place in Seizure Detection Challenge [leaderboard]

  Organized by The International Conference on Artificial Intelligence in Epilepsy and Other Neurological Disorders 2025

• CICS Donor-Funded Scholarships

  Funded by the Manning College of Information & Computer Sciences (CICS) at UMass Amherst

Technical Contribution

China Construction Bank Achievement System

React.JS, Go, MySQL, Azure

China Construction Bank is one of China's four major state-owned banks. The Dianlong Branch in Kunming is a secondary branch of the China Construction Bank, with 11 sub-branches in Kunming and hundreds of employees. To help the branch managers track and manage the performance of each sub-branch, position, and employee, I implemented a web-based performance tracking system. Employees can submit their daily performance metrics, such as the amount of fixed deposits, current deposits, financial products sold, and the number of new accounts opened for customers. Managers can view performance scores calculated based on rules defined by the bank and systematically evaluate the performance of each employee. Comprehensive Restful APIs, are provided for managers to sort and filter the data. The system is used by all 11 sub-branches in Kunming city from 2024-2025.

Submitty Open Source

HTML, JavaScript, PHP, SQL

Submitty is an open source course management, assignment submission, exam and grading system that is widely used by Computer Science departments in Rensselaer Polytechnic Institute and other universities. I contributed to the development of the system as a full-stack developer in 2023 spring by implementing several new features like customizable pronouns setting, which involved front-end development using HTML and JavaScript, backend API creation with PHP, and database schema design using SQL. I also found and fixed several bugs in the system. So far, 5 of my pull requests have been merged into the main branch of the Submitty repository.

Wox Translater Plugin

Python, BeautifulSoup

Wox is a full-featured launcher, access programs and web contents as you type. I developed a plugin for Wox that can translate English to Chinese and vice versa by scraping Cambridge dictionary using BeautifulSoup. Because of the nature of scraping, users can directly download the plugin to use without any registration or API key. The plugin is currenly live on the Wox plugin store.

Ocean Dynamic Wallpaper

HTML, JavaScript

This is a dynamic wallpaper that simulates the ocean written in Canvas element in HTML and animation in JavaScript. Bubbles are generated with random radius, color, and moving speed, where users can click bubbles by tracking the mouse's coordinates. Project repository can be found in GitHub.