Research

My research develops machine learning methods for reasoning about complex social and urban systems, particularly when spatial structure, temporal evolution, and network interactions jointly shape system behavior. These are domains where classical models fail and where deep learning must encode constraints from geography, infrastructure, and human interaction networks.

I work on applications involving cities, elections, epidemics, and mobility patterns, systems characterized by high spatial dimensionality, non-stationary temporal signals, and dynamic network topologies. My goal is to integrate spatio-temporal deep learning with computational social science to produce models that are both methodologically rigorous and empirically relevant for urban science, political forecasting, and public health.

Research Areas

Spatio-Temporal Machine Learning

Developing neural architectures that respect the structure of geographic space and continuous time. My work combines graph neural networks with temporal modeling approaches (point processes, neural ODEs) to enable prediction on irregular spatial networks with heterogeneous temporal dynamics.

Research Questions

  • •How do we design multi-scale representations for city-level spatial prediction?
  • •What are the right inductive biases for learning on geographic networks?
  • •Can we transfer knowledge across geographies with different data availability?

Network Science & Graph Learning

Studying how network structure influences dynamical processes from disease transmission on mobility networks to information diffusion in social graphs. I develop scalable GNN algorithms that can handle billion-edge spatial networks while maintaining theoretical guarantees.

Research Questions

  • •How do we scale graph learning to massive geographic networks?
  • •What graph properties enable or hinder learning in spatial domains?
  • •How should we handle non-stationary distributions across network regions?

Computational Social Systems

Understanding how individual behaviors aggregate into population-level outcomes through networked interactions. Applications include election forecasting from social media, human mobility modeling, and emergent coordination in multi-agent systems.

Research Questions

  • •How do we model information cascades that drive electoral outcomes?
  • •What role does network structure play in shaping mobility patterns?
  • •Can we design ML systems that provide transparent uncertainty quantification for social predictions?

Urban Growth & Infrastructure

Predicting urban expansion patterns and infrastructure needs using satellite imagery, demographic data, and transportation networks. My methods combine computer vision with structured spatial prediction to support sustainable city planning.

Research Questions

  • •How do we integrate satellite imagery with graph-structured urban data?
  • •What temporal resolutions are needed to capture urban change processes?
  • •How can models account for policy interventions and planning decisions?

Causal Inference in Spatial Data

Developing methods for causal identification when units are embedded in space or networks. I work on approaches that handle spatial interference, spillover effects, and latent geographic confounding using deep generative models.

Research Questions

  • •How do we identify causal effects under spatial interference?
  • •Can neural networks help discover latent spatial confounders?
  • •What sensitivity analysis methods work for unmeasured spatial confounding?

Research Philosophy

Domain-grounded ML: I collaborate closely with domain experts to ensure models are scientifically meaningful and address real problems in urban planning, epidemiology, and political science.

Theoretical rigor: I prove when methods work, establishing sample complexity bounds, identifiability conditions, and generalization guarantees.

Scalability: I build systems that scale to real-world data: billions of edges, millions of time points using efficient sampling strategies and distributed computing.

Interpretability: I develop models whose predictions can be explained to stakeholders through attention mechanisms, feature attribution, and uncertainty quantification.

Social responsibility: I consider fairness, privacy, and ethical implications of spatial prediction systems, particularly their potential to amplify existing inequalities.