ANNOUNCEMENT OF DEFENSE OF DISSERTATION RESEARCH
The faculty of the Engineering Management and Systems Engineering department is pleased to issue an invitation to Ms. Niharika Deshpande鈥檚 defense of the research conducted for her dissertation.
The defense is open to the public.
Date: Monday, March 23rd, 2026
Time: 4:00 PM to 5:30 PM
Location: Engineering Systems Building, EMSE Conference Room 2101A
Online Access: Via Zoom Meeting ID: 938 0234 5920 Passcode: 948639
Doctoral Candidate: Niharika Deshpande
Physics-Guided Deep Learning for Predictive Modeling of
Spatiotemporal Dynamical Systems
Director: Dr. Hyoshin Park
Abstract:
Many physical and networked systems operate under complex spatial and temporal variability. Transportation networks respond to fluctuating demand, atmospheric fields reorganize as storms intensify, and coastal response depends on localized forcing pathways. Modeling such systems requires learning frameworks that operate on irregular geometries, capture spatiotemporal dependencies, and provide interpretable measures of predictive uncertainty.
This dissertation develops a physics-guided spatiotemporal learning framework for structured dynamical systems whose governing interactions are neither static nor Euclidean. The central premise is that spatial relationships are geometry-dependent and must be learned directly on the underlying network structure. To represent this behavior, system states are modeled on time-varying graphs whose connectivity encodes localized spatial interactions and directional dependencies among system components. This formulation enables learning on non-Euclidean domains while preserving physically meaningful spatial structure.
Temporal dynamics are integrated within the graph representation so that spatial organization and sequential behavior are learned jointly. The framework further incorporates probabilistic latent representations that quantify predictive uncertainty and enable assessment of forecast reliability and calibration.
The methodology is evaluated in multiple structured dynamical contexts, including traffic state estimation and hurricane-driven storm surge prediction. Across these settings, the framework produces spatially coherent field predictions, stable performance across varying conditions, and uncertainty estimates that scale systematically with system intensity.
This work establishes a general methodology for uncertainty-aware spatiotemporal prediction on network-constrained domains. By embedding dynamic spatial structure and physically guided learning within graph-based models, the dissertation advances the development of adaptable forecasting systems for complex environmental and engineered processes.
Bio:
Niharika Deshpande is a Ph.D. candidate in Engineering Management and Systems Engineering at 国产伦理, where she works as a Graduate Research Assistant. Her research focuses on machine learning for complex spatiotemporal systems, including applications in transportation networks and hurricane-driven storm surge prediction. She holds a Master鈥檚 degree in Applied Physics from the Technical University of Munich, Germany, and a Bachelor of Technology in Engineering Physics from the Indian Institute of Technology Guwahati, India.