About

Professor Semiha Ergan is recognized for her contributions to urban science and education at NYU Tandon School of Engineering. She has been awarded the University Distinguished Teaching Award, highlighting her excellence in teaching. Her work is associated with the Center for Urban Science + Progress (CUSP), an interdisciplinary center dedicated to applying science, technology, engineering, math, and social sciences to urban communities. While the specific details of her research focus are not explicitly described in the provided text, her recognition and involvement with CUSP suggest her engagement in urban data science, innovative urban solutions, and possibly in areas related to urban health, infrastructure, and environmental concerns. Her contributions are part of a broader effort to develop data-driven approaches that improve urban quality of life and address socioeconomic, environmental, and infrastructural challenges.

Research topics

• Computer Science
• Artificial Intelligence
• Machine Learning
• Computer vision
• Psychology
• Human–computer interaction
• Cartography
• Engineering
• Geography

Selected publications

• An ontology-based approach for Façades inspection checklist generation

  Advanced Engineering Informatics · 2026-02-16

  articleSenior authorCorresponding
  PublisherDOI

• Window view optimization in building façade layout design (ViewOpt): A Human-centric 3D simulation and computer vision framework

  Building and Environment · 2026-03-22

  articleSenior authorCorresponding
  PublisherDOI

• Leveraging Large Language Models (LLM) and Machine Learning for Extracting Conditions that Impact Safety Rating of Building Façades

  2026-01-28

  articleSenior author

  Defective building façades pose significant safety risks and require regular inspection to prevent accidents and ensure public safety. One challenge faced is the lack of guidelines to help inspectors rate the condition of façades given the identified defects. Current means fail to help inspectors determine how defect types, materials, and façade components relate to safety ratings. This study utilizes large language models (LLMs) and machine learning methods to analyze historical façade inspection reports to extract relationships between defect types and their characteristics, façade materials, and façade components to assess their impact on assigned safety ratings, serving as the basis for such a needed guideline for consistent risk rating. This approach also reveals patterns in conditions and components that are most prone to unsafe states. We also demonstrated that building components play the most significant role in determining safety risks, surpassing defect and material types. Additionally, cracks in terra cotta ornamentation were uniquely linked to unsafe states due to their high fall risk. These findings help to uncover the reasoning behind safety ratings, contributing to the development of standardized rating guidelines to reduce discrepancies and ensure consistent risk assessments across inspections.

  PublisherDOI

• A Taxonomy of Urban Façade Defects and Their Distribution on Façade Components: A Data-Driven Analysis of Historical Inspection Reports

  Journal of Construction Engineering and Management · 2025-07-28 · 5 citations

  articleSenior author

  Well-maintained building façades are essential for public safety. Major cities across the globe have inspection programs to ensure façade safety. However, accidents caused by falling debris and complaints continuously happen and indicate a need to improve the current inspection practices. A systematic guideline is lacking to achieve comprehensive safety inspections of buildings. A challenge that exists on the path toward a systematic and comprehensive façade inspection process is the lack of availability of common and comprehensive vocabularies for the façade safety inspection domain that extends beyond material-based defects. This research defines a taxonomy of façade defects and relationships of these defects to façade component hierarchies, through natural language processing and machine learning methods employed on more than 2,500 façade inspection reports for three major façade types. Contributions of this work include the taxonomy of defect types and their applicability to façade components, forming the missing but foundational piece for enabling comprehensive inspection of building façades. The study can directly be used in practice to establish a standard inspection practice across various inspection companies and related city agencies.

  PublisherDOI

• STRAP: Spatial-Temporal Risk-Attentive Vehicle Trajectory Prediction for Autonomous Driving

  2025-11-18

  articleSenior author

  Accurate vehicle trajectory prediction is essential for ensuring safety and efficiency in fully autonomous driving systems. While existing methods primarily focus on modeling observed motion patterns and interactions with other vehicles, they often neglect the potential risks posed by the uncertain or aggressive behaviors of surrounding vehicles. In this paper, we propose a novel spatial-temporal risk-attentive trajectory prediction framework that incorporates a risk potential field to assess perceived risks arising from behaviors of nearby vehicles. The framework leverages a spatial-temporal encoder and a risk-attentive feature fusion decoder to embed the risk potential field into the extracted spatial-temporal feature representations for trajectory prediction. A risk-scaled loss function is further designed to improve the prediction accuracy of high-risk scenarios, such as short relative spacing. Experiments on the widely used NGSIM and HighD datasets demonstrate that our method reduces average prediction errors by 4.8% and 31.2% respectively compared to state-of-the-art approaches, especially in high-risk scenarios. The proposed framework provides interpretable, risk-aware predictions, contributing to more robust decision-making for autonomous driving systems.

  PublisherDOI

• Sim-to-Real Transfer Learning for Large-Scale Short-Term Building Energy Forecasting in Sustainable Cities

  2025-11-11

  articleOpen accessSenior author

  Short-term building energy forecasting is essential for optimizing operations, enabling demand response, and improving network reliability. However, its large-scale deployment across building portfolios remains limited, constraining both cost savings and decarbonization efforts. Although data-driven models often outperform physics-based and reduced-order methods in terms of operational efficiency and accuracy, they generally rely on extensive, high-quality historical real-world building, system, and usage data that most buildings do not possess. Transfer learning (TL) mitigates data scarcity by transferring knowledge from data-rich "source" buildings to data-poor "target" buildings. However, publicly available datasets rarely contain a sufficiently large and diverse set of real buildings to support portfolio-scale modeling. As a result, existing real-to-real TL approaches rely on per-building similarity matching and fine-tuning. We introduce STARS (Synthetic-to-real Transfer for At-scale Robust Short-term forecasting), a two-stage pipeline that generates portfolio-level, zero-shot, day-ahead electricity-load predictions for buildings. Stage 1 trains a lightweight variational autoencoder (VAE) on 8,856 hourly load profiles from the New York State (NYS) subset of the ComStock simulation library and, by evaluating reconstruction error over a calendar-aligned two-week window, retains real buildings whose load behavior is well captured by these synthetic models. Stage 2 applies a Patch-based Time-Series Transformer (PatchTST), pre-trained on the same NYS synthetic corpus, to generate day-ahead forecasts for all buildings that pass the screen. Evaluated on 101 metered buildings from the NYS subset of the Building Data Genome Project 2, STARS attains a mean CVRMSE of 12.07% in the summer months over 83 VAE screened targets (82.2% coverage) and 11.44% in the winter months over 85 VAE screened targets (84.2% coverage), both well below the well-calibrated threshold 30% in ASHRAE Guideline 14. This level of accuracy supports reliable portfolio-scale load prediction that can advance grid-efficiency and decarbonization goals.

  PublisherOA PDFDOI

• Quantification of the impact of street design features on restorative quality in urban settings

  Sustainable Cities and Society · 2025-02-13 · 12 citations

  articleSenior author
  PublisherDOI

• VR-WISE: VR based work zone immersive simulation with eye tracking and biometric sensors for capturing driver’s awareness

  2025-09-21 · 1 citations

  article
  PublisherDOI

• Evaluating GAN Architecture for Generating Images of Defective Façades

  2025-12-11 · 1 citations

  articleSenior authorCorresponding

  Building façade inspections are of paramount importance for public safety, as defect-related falls have led to severe accidents over the years. Traditional, expertise-reliant manual methods are laborious, time-consuming, and unsafe due to height-related risks. Consequently, there has been a growing interest in employing computer vision, especially deep learning (DL), techniques for automated defect detection. However, the effectiveness of DL models is significantly hampered by a shortage of extensive, annotated defect data sets. While collecting and annotating façade defect images contributes to data scarcity and the varying frequency of occurrence of defects creates an imbalance in data sets, this is concerning because such infrequent defects still pose significant safety risks. This study tackles these data imbalance and scarcity issues by deploying Generative Adversarial Network (GAN) algorithms to create synthetic images of façades with defects. Multiple GAN algorithms have been evaluated using Fréchet Inception Distance (FID) and visual evaluations. Outperforming the GAN algorithm is suggested to furnish a balanced training data set for façade defect identification models.

  PublisherDOI

• Spatial clustering and NLP-based analysis of defect patterns on urban façades: Implementation on NYC buildings

  Cities · 2025-07-01 · 1 citations

  articleSenior authorCorresponding
  PublisherDOI

Recent grants

• FMSG: ARM4MOD: AI-powered and Robot-assisted Manufacturing for Modular Construction

  NSF · $500k · 2021–2025

Frequent coauthors

• Reza Akhavian

  San Diego State University

  36 shared

• Fei Dai

  36 shared

• Burcu Akinci

  Carnegie Mellon University

  28 shared

• Amir H. Behzadan

  Texas A&M University

  25 shared

• Zhengbo Zou

  University of British Columbia

  21 shared

• Jing Du

  University of Florida

  20 shared

• Eric Du

  West Virginia University

  16 shared

• Xinran Yu

  Guangzhou Institutes of Biomedicine and Health

  12 shared

Labs

• Building Informatics and Visualization LabPI

  We innovate solutions for the operational challenges associated with construction and operation of facilities and infrastructure systems in urban settings.

Awards & honors

• DARPA Young Faculty Award (2015)
• Berkman Faculty Award, Carnegie Mellon University (2013)
• Outstanding Teaching Assistant Award, Carnegie Mellon Univer…
• Winner of Association for Advancement of Cost Engineering (A…
• Higher Education Council Doctoral Student Scholarship (2003-…