From methods to applications
Projects
Current research and selected earlier work across graph learning, AI systems, and scientific applications.
Graph learning & system design
Published & ongoing
Scalable graph sampling & sparsification
From attribute-guided sampling to learned sparsifiers, support graphs, and edge-disjoint decomposition: develop sparse graph representations that preserve useful structure while reducing GNN computation and memory.
Preprint & ongoing
GNN unification & architecture generation
Use a common layer equation to compare graph neural network families and explore structurally consistent architecture generation through propagation, message passing, aggregation, and transformation.
HPEC 2026 · accepted
Joint hardware/software partitioning & scheduling
Develop differentiable GNN and reinforcement learning methods for coupled partitioning and scheduling decisions in system design, accounting for resource, area, and communication constraints.
Ongoing
Graph-based circuit equivalence checking
Investigate GNN-based subcircuit identification, gate embeddings, and graph matching to support circuit equivalence verification. Current research builds on graph representations of circuit structure.
Ongoing
Graph sparsification methods & benchmarks
Develop a comprehensive textbook and benchmarking framework spanning graph sparsification, rewiring, augmentation, theoretical guarantees, and applications. Compare methods across graph types and downstream learning tasks.
Document intelligence & agentic AI
Public dataset
NEPATEC 2.0
Contributed to a standardized corpus of NEPA documents and metadata for research in environmental permitting and document intelligence. The public dataset connects project, process, document, and page-level information.
Ongoing
NEPATEC 2.5: multimodal metadata extraction
Develop multimodal LLM methods to extract metadata from heterogeneous NEPA documents, geolocate projects from map images, and generate project summaries. This is ongoing work described in the October 2026 CV.
Ongoing
PermitCE: AI-assisted document search & chat
Build a retrieval-augmented system for NEPA categorical exclusion documents, supporting semantic search, recommendations, summarization, and interactive question answering.
Cybersecurity & applied machine learning
Published & ongoing
Vulnerability mapping & threat intelligence
Connect CVEs with weakness classes and attack patterns using language models, and integrate security repositories into knowledge graphs for multi-hop analysis. Study scalability on modern AI accelerators.
Ongoing
Composite metric for cyber risk prioritization
Develop a unified vulnerability prioritization metric that combines CVSS, CWSS, EPSS, and LEV to bring severity, weakness context, exploit likelihood, and exploitation evidence into a common assessment.
Collaborative research
Predictive analytics for russet potato selection
Apply machine learning, feature selection, and predictive modeling to phenotypic data to support decision-making in potato breeding.
Collaborative research
Machine learning for the USEPA SPECIATE database
Apply clustering to chemical source profiles to identify similarities in species distributions and support environmental data exploration.
Earlier work
Published
Multi-objective & many-objective optimization
Study multi-objective and many-objective evolutionary optimization, including adaptive fuzzy dominance, reference-point guidance, and the balance between Pareto convergence and solution diversity. Apply these methods to cluster performance and scheduling with competing objectives.
Project descriptions and status reflect the October 2026 CV. Resources are linked where a public paper, implementation, or dataset is available.