Zirou Qiu

Ph.D. Student in CS @ UVA; CV; Google Scholar

I am a Ph.D. student in Computer Science at the University of Virginia working with Madhav Marathe. I am also (i) a research assistant at the Biocomplexity Institute, working with S. S. Ravi, Dan Rosenkrantz, Dick Stearns; and (ii) a reseach intern at Fujitsu Research of America.

I study the foundations of large complex systems (e.g., social systems, multi-agent systems, logistics, infrastructure and transportation systems) represented as graphs using techniques in machine learning, optimization, and computational modeling. I focus on developing formal models and algorithms to simulate, learn and optimize these large systems, aiming to improve system performance, understand system behaviors, and benefit the social good.

Research Interests

  • Machine Learning

  • Optimization

  • Computational Modeling

  • Computational Social Science & Epidemiology

  • Game Theory

  • Mathematical Programming

  • Network Science


Publications / Preprints

You may click the title of a paper to view its pdf file.

Welfare Optimization for Resource Allocation with Peer Effects

In Submission

Zirou Qiu, Daniel J. Rosenkrantz, Matthew O. Jackson, Simon A.Levin, S. S. Ravi, Richard Stearns, Madhav Marathe

Summary. We developed a principled model for resource allocation with externalities (e.g., mutual influences among the resources or the population) in complex systems. We proposed efficient strategies that guarantees optimal or near-optimal welfare outcomes for real-world re- source allocation scenarios

Keywords. Resource allocation, optimization, large systems planning, computational social science

Learning Discrete Dynamical Systems under Classification Noise

In Submission

Zirou Qiu*, Zakaria Mehrab*, Abhijin Adiga, Madhav Marathe, S.S. Ravi, Daniel Rosenkrantz, Richard Stearns and Anil Vullikanti.

Summary. We proposed efficient noise-tolerant algorithms for learning the behaviors of large-scale systems in realistic noisy environments, with a provable high learning accuracy. Our algorithms are highly scalable to large networks, and the amount of training data required is provable near-optimal.

Keywords. Machine learning on graphs; learing under noise; complex systems; sample complexity

Theoretical Foundations for Parent Divorcing Transformations in Bayesian Networks

In Submission

With Daniel Rosenkrantz, Madhav V. Marathe and S. S. Ravi

We estabilished a theoretical groundwork for dimensionality reduction in Bayesian inference. Further, we proposed efficient algorithms for dimensionality reduction with provable guarantees.

Keywords. Bayesian Networks; Parent divorcing transformation; Network embedding; Theoretical foundations

Efficient PAC Learnability of Dynamical Systems Over Multilayer Networks

International Conference on Machine Learning (ICML-2024) [Acc.rate: 27.5%]

Zirou Qiu, Abhijin Adiga, Madhav Marathe, S.S. Ravi, Daniel Rosenkrantz, Richard Stearns and Anil Vullikanti.

We proposed the first efficient learning method for inferring the behaviors of large-scale multilayer social systems and multi-agent systems. Further, we lay a theoretical foundation of learning multilayer systems with an exact analysis of the model complexity and expressiveness

Keywords. Machine learning on graphs; multilayer graphs; complex systems; sample complexity

Learning the Topology and Behavior of Discrete Dynamical Systems

AAAI Conference on Artificial Intelligence (AAAI-2024) [Acc.rate: 23.8%]

Zirou Qiu, Abhijin Adiga, Madhav Marathe, S.S. Ravi, Daniel Rosenkrantz, Richard Stearns and Anil Vullikanti.

We investigate the core principles of learning both the behaviors and the topology of black-box social systems and multi-agent systems. Towards this end, we proposed several efficient learning algorithms with provable guarantees for realistic scenarios of learning large systems.

Keywords. Machine learning on graphs; complex systems; sample complexity;

Assigning Agents to Increase Network-Based Neighborhood Diversity

Intl. Conf. Autonomous Agents and Multiagent Systems (AAMAS-2023) [Acc.rate: 23.3%, Oral]

Zirou Qiu, Andrew Yuan, Chen Chen, Madhav Marathe, S.S. Ravi, Daniel Rosenkrantz, Richard Stearns and Anil Vullikanti.

We developed a formal model for real-world spatial resources allocation problems, such as public housing assignment and urban spatial planning. Further, we presented rigorous methodologies for resource allocation with provable close-to-optimal guarantees, while achieving social diversity and fairness.

Keywords. Resource allocation and planning; optimization; fairness and diversity

Networked Anti-Coordination Games Meet Graphical Dynamical Systems: Equilibria and Convergence

AAAI Conference on Artificial Intelligence (AAAI-2023) [Acc.rate: 19.6%, Oral]

Zirou Qiu, Chen Chen, Madhav Marathe, S.S. Ravi, Daniel Rosenkrantz, Richard Stearns, Anil Vullikanti.

We presented a theoretical groundwork for stability and convergence in real-word strategic situations such as resources sharing and competition within a society.

Keywords. Game theory; Nash equilibrium, convergence time

Airborne Disease Transmission During Indoor Gatherings over Multiple Time Scales: Modeling Framework and Policy Implications

Proceedings of the National Academy of Sciences (PNAS-2023)

Avinash Dixit, Baltazar Espinoza, Zirou Qiu, Anil Vullikanti, and Madhav Marathe

We introduced a principled modeling framework that couples the fast dynamics of viral loads in the local scale and the slow dynamics of disease progression at the population level. Further, we derived policy guidelines to lessen the negative impact of epidemics.

Keywords. Group gathering; multiple time scales; large-scale modeling; pandemic science;

Understanding the Coevolution of Mask Wearing and Epidemics:
A Network Perspective

Proceedings of the National Academy of Sciences (PNAS-2022)

Zirou Qiu, Baltazar Espinoza, Vítor V. Vasconcelos, Chen Chen, Sara M. Constantino, Stefani A. Crabtree, Luojun Yang, Anil Vullikanti, Jiangzhuo Chen, Jörgen Weibull, Kaushik Basu, Avinash Dixit, Simon Levin, Madhav Marathe.

We presented a fundamental model for dueling dynamics of social and biological contagions both cooperating and completing in a population. Through our modeling, we uncovered a collection of previously unknown phenomena in contagion dynamics. We derived corresponding policy guidelines to suppress epidemics and prevent their resurgence.

Keywords. Dueling dynamics, multilayer networks, non-pharmaceutical interventions, social science

Finding Nontrivial Minimum Fixed Points in Discrete Dynamical Systems: Complexity, Special Case Algorithms and Heuristics

AAAI Conference on Artificial Intelligence (AAAI-2022) [Acc.rate: 15%, Oral: 4.8%]

Zirou Qiu, Chen Chen, Madhav Marathe, S.S. Ravi, Daniel Rosenkrantz, Richard Stearns, Anil Vullikanti.

We established the principles of finding equilibrium points in contagion diffusion within a population, with the goal of minimizing the overall scale of infection in a society.

Keywords. Optimization; complex systems; fixed point; contagion diffusion

Efficiently Learning the Topology and Behavior of a Networked Dynamical System Via Active Queries

International Conference on Machine Learning (ICML-2022) [Acc.rate: 22%, Spotlight]

Daniel Rosenkrantz,(α-β)Abhijin Adiga, Madhav Marathe, Zirou Qiu, S.S. Ravi, Richard Stearns, Anil Vullikanti

We presented the first collection of active learning based methods for inferring both the behaviors and the topology of black-box social systems and multi-agent systems. Further, we conducted an extensive experimental study on the empirical performs on the proposed algorithms on real-world networks over various learning scenarios.

Keywords. Machine learning on graphs; Active learning; query model; complex systems; query complexity

ELRUNA: Elimination Rule-based Network Alignment

ACM Journal of Experimental Algorithmics (ACM-JEA-2021)

Zirou Qiu, Ruslan Shaydulin, Xiaoyuan Liu, Yuri Alexeev, Christopher S. Henry, Ilya Safro

Summary. We propose novel algorithms for the network alignment problem that outperforms the state-of-the-art methods, with applications in computational biology and social network analysis.

Keywords. Optimization; network alignment; social network analysis


Research Experience

  • Working in a cross-disciplinary team with over 40 scientists from fields such as CS, social science, and economics. Developed communication, project management, and teamwork skills within a large group.

  • Researching the foundations of large-scale systems (e.g., social, multi-agent, infrastructure, and transportation systems) via machine learning, optimization and mathematical modeling.

  • Developing large-scale frameworks for socio-technical system modeling and data analysis.

  • Being the lead researcher in four research projects, with publications at various venues (PNAS 22 & 23, AAMAS 23, AAAI 22, 23 & 24, ICML 22 & 24). Made important contribution to NSF Expeditions Project in Computing.

Fall 2020 - Present

  • Studied computational problems in quantum computing and biology.

  • Built a pipeline for microbiome modeling. Discovered novel microbiome migration pattern.

Jan 2019 - May 2020

  • Investigated optimization problems in large networks and developed novel metaheuristics to solve the network matching problem. The proposed algorithms achieved over 2 times performance improvement compared to state-of-the-art methods. Published as first author at ACM-JEA.

Summer 2019


Other Stuff

What is a discrete dynamical system?

My research focuses on the foundation of complex systems, although one won't often find the term "complex systems" in my papers. Instead, you may note that my work involves things called discrete dynamical systems.

In essence, the discrete dynamical system is a mathematical framework that models complex systems represented as graphs. Importantly, discrete dynamical systems provide a theoretical foundation to formally address questions about complex systems. Such questions include (1) convergence and stability of system dynamics; (2) learning black-box systems; (3) resource allocation and planning; (4) large-scale (agent-based) simulation.

Overall, I investigate complex systems through the structured framework of discrete dynamical systems. Desipte this connection, the problems, techniques, and results obtained in my work apply to much border realms of real-world systems (e.g., social, multi-agent, and infrastructure systems).