Projects
Data-driven assessment & optimization of life-cycle inspection and maintenance decisions under uncertainty
This PhD research focuses on the development of computational frameworks that incorporate time-varying risk & reliability concepts in optimal design, monitoring, and maintenance of structural systems. This includes data-driven assessment and optimization enabled by probabilistic inference, structural reliability, and reinforcement learning methods.
Learning to disentangle dependencies in multi-component structural systems for scalable decisions in high-dimensional spaces
This PhD research approaches built environment decisions from a system-level perspective, aiming at disentangling the inherent high-dimensional correlations and dependencies that exist in data and models of engineering networks. It provides insights in questions related to the value of information, joint model updating, and coordination of autonomous agents acting on distributed components through Bayesian networks and decentralized reinforcement learning.
Deep Visual Similarity Learning for Architectural Drawings
Architects communicate their designs through various visual abstractions of the physical space; including orthographic drawings, photos, and 3D models. Semantic similarity learning for architectural drawings is a PhD project of Casper van Engelenburg that started in October 2021, focusing on understanding visual patterns in floorplan image data. He develops deep contrastive learning frameworks that enables us to learn low-dimensional, task-agnostic representations of architectural drawings. This research line builds a foundation for large quantitative analysis of archival and linked visual data.
Beyond Appearance in Architectural Visual Data
At the confluence of computer vision, data science, and architecture, a building is a high-dimensional data instance, predominantly represented in visual computer models encoding both aesthetic and performance indexes. Fatemeh Mostafavi started her PhD in June 2022 and focuses on performance-driven abstraction and generation of design models using computer vision. She is developing deep learning models that are able to make sense of a building plan with respect to sustainability, daylight, and comfort.