Yanhai Xiong

This research project explores complex microbial communities in diverse environments, such as treated wastewater and animal digestive systems. While a single microbial community can host thousands of unique species—each carrying vast genetic data in its DNA—only a subset of those genes are actively expressed as RNA under specific environmental conditions.

Our long-term objective is to deploy computational tools and machine-learning frameworks to analyze and predict gene expression patterns. Specifically, we aim to model which genes are expressed, by which species, and at what specific levels, using genomic and environmental datasets.

Student Responsibilities & Qualifications:

• Students will focus primarily on the data science, programming, and computational modeling dimensions of the project.
• Prior background in biology or bioinformatics is highly advantageous but not strictly required.
• Applicants must demonstrate a strong comfort level with independent learning and grasping new technical concepts.

To better understand the scope and methodology of this research, please review the foundational study: Tang & Lau (2025), “Varying effects of chlorination on microbial functional repertoire and gene expression in contrasting effluents,” Frontiers in Microbiology.

Our research focuses on developing intelligent decision-making methods for complex and dynamic environments involving multiple interacting agents. We integrate game theory, reinforcement learning, combinatorial optimization, and neural planning to address challenges in strategic interaction, cooperation, sequential decision making, and uncertainty. Our recent work explores learning-based approaches to stochastic combinatorial optimization, hybrid neural optimization, and multi-agent path finding under partial observability. These methods have applications in autonomous systems, resource allocation, cybersecurity, and intelligent infrastructure.

Papers:

• Multi-Agent Cooperation and Reinforcement Learning: HogRider: Champion Agent of Microsoft Malmo Collaborative AI Challenge, AAAI'18 [pdf]
• LENS: Learning to Navigate with Active Search for Partially Observable MAPF in Unknown Environments, TMLR'26 [pdf]
• HyCO: A Hybrid Neural Solver for Combinatorial Optimization, NeurIPS'26 [pdf]

Android is the most widely used operating system. As a result, it is also favored by numerous attackers. Attackers develop Android malware to perform malicious behaviors on users' Android devices for stealing sensitive information or money. Part of my work is using machine learning and optimization to detect and analyze Android malware.

Papers:

• Detection via (Somewhat) Robust Irreversible Feature [pdf]
• FAMCF: A few-shot Android malware family classification framework [web]
• TraceRAG: A LLM-Based Framework for Explainable Android Malware Detection and Behavior Analysis, Preprint [pdf]

Book:

• The Android malware handbook: detection and analysis by human and machine [web]

Papers on information security:

• Strategic vulnerability exploit and disclosure in cyber walfare, IEEE SJ'20[pdf]

According to Symantec, there average gap from the time a company is compromised by a zero-day attack to the time the vulnerability is discovered is 312 days. This leaves an adversary with a lot of time to exfiltrate corporate IP. Recent work has suggested automatically generating multiple fake versions of a document to impose costs on the attacker who needs to correctly identify the original document from a set of mostly fake documents.

Papers:

• Realistic Fake Equation Generation [pdf]

As a promising solution of clean energy transportation, electric vehicles (EVs) have drawn interests from different communities. For successful introduction of EVs, the construction of charging facilities is of top-priority since it is essential to mitigate drivers' anxiety of running out electricity. We study the placement and management of EV charging stations with game-theoretical approaches, in which the mutual influence between EV infrastructure and the environment is delicately considered, especially with the consideration of the strategic charging behavior of participating human subjects, i.e., EV users.

Papers:

• EV Charging Station Placement, IJCAI'15 [pdf], Extended journal version, IEEE-TITS'17[pdf]
• EV Charging Station Management, AAMAS'16 [pdf]
• Complete version, PhD Thesis'18 [pdf]
• Charging Strategy Study and the Application, JAAMAS'20 [pdf]
• Planning, Scheduling, and Behavior in EV Charging Systems: A Critical Survey and Trilemma Framework, Preprint [pdf]

Another work based on our champion agent which won 2017 Microsoft Malmo Collaborative AI Challenge is potential to be helpful in future research, expecially when auto-drive vehicles come to reality and. The Microsoft Malmo Collaborative AI Challenge (MCAC), which is designed to encourage research relating to various problems in Collaborative AI, builds on a Minecraft mini-game called “Pig Chase”. Pig Chase is played on a 9 × 9 grid where agents can either work together to catch the pig and achieve high scores, or give up cooperation and achieve low scores. After playing certain episodes (e.g., 100) of games, the agent who achieves the highest average scores wins the challenge.

The solutions underlying our agent HogRider are characterized by

• A novel agent type hypothesis approach for dealing uncertainties about the type of the other agent and the observation of the other agents' actions.
• A novel Q-learning framework to learn an optimal policy against each type of agent, which 1) employs state-action abstraction to reduce state-action space, 2) utilizes warm-start of Q-functions with pre-traning, and 3) exploits active-ε-greedy for active exploration of potentially more beneficial actions.

Paper and media coverage:

• Hogrider, AAAI'18 [pdf]
• SCSE Team HogRider wins first place for the Microsoft Azure for Research grant prize! [link]
• (News in Chinese) 微软Malmo项目协同AI挑战赛：中国团队榜上有名，获胜者福利多多 [link]