Topological and Geometric Machine Learning
I develop learning methods that use topological and geometric information as explicit components of the representation or model architecture. This work studies how connectivity, cycles, multiscale structure, and geometry can complement node features and conventional graph statistics.
Representative methods include topology-aware transformers, topological tokenization, and multiparameter topological descriptors for structured scientific data.
Graph Representation Learning
My graph-learning research focuses on node classification, graph classification, and relational representation learning. I study how spatial context, class geometry, positional information, and supervision-derived coordinates can improve graph neural networks, transformers, and feature-based models.
Temporal and Spatio-Temporal Learning
I investigate temporal graph classification and learning from evolving networks. The goal is to combine structural, topological, spectral, and temporal signals without discarding changes that are important for prediction and anomaly detection.
AI for Power-Grid Resilience and Cybersecurity
My applied research develops machine-learning tools for outage detection, cyberattack detection, anomaly localization, and resilient monitoring in power systems. Applications include grid-connected photovoltaic systems, electric-vehicle charging infrastructure, and large distribution networks.
Selected Research Projects
TopoFormer
A topology-aware attention framework for graph learning that represents structural information in a form suitable for transformer-based models.
T3former
A temporal graph classification framework that combines graph, topological, and spectral information across time.
SCNode
Spatial and contextual coordinates for graph representation learning, designed to complement conventional node features and graph structure.
MP-Grid
A topological machine-learning approach for detecting outages in power-distribution systems using multiparameter information.