Research
I'm interested in natural language processing, computer vision, multimodal learning, deep learning, generative AI, and human-computer interaction. Most of my research is about inferring the knowledge (i.e., best practices) of construction planning and controls from textual (e.g., schedules, daily reports, change orders, RFI, etc.) and visual (e.g., images, point clouds, etc.) information. Representative papers are highlighted.
Journal Papers
Yoonhwa Jung, Ikhyun Cho, Shun-Hsiang Hsu, Mani Golparvar-Fard
Automation in Construction, 2024
project page / video / Journal paper
A Vision Transformer-based image captioning model, VisualSiteDiary, which creates human-readable captions for daily progress and work activity log, and enhances image retrieval tasks. Present a new image-caption pair dataset (VSD dataset) and Demo toward a real-time construction site daily log reporting. Superior-quality captions are generated compared to the state-of-the-art image captioning models.
Yoonhwa Jung, Julia Hochenmaier, Mani Golparvar-Fard
Automation in Construction, 2024
Journal paper / Conference paper
A new NLP transformer model, UniformatBridge, to automatically map schedule activities to ASTM UniFormat classes, embedding construction schedule sequencing knowledge. UniformatBridge serves as a universal identifier enabling automated creation of 4D BIMs and streamlines mapping between schedule, cost and payment application data.
Fouad Amer, Yoonhwa Jung, Mani Golparvar-Fard
Automation in Construction, 2023
Journal paper
A NLP transformer method for deciphering implicit construction dependencies with the role and flexibility of activity relationships. A method for revising lookahead plans based on the flexibility of activity dependencies, mitigating the risk of delays.
Fouad Amer, Yoonhwa Jung, Mani Golparvar-Fard
Automation in Construction, 2021
Journal paper
This paper presents the first attempt to automate linking look-ahead planning tasks to master-schedule activities following an NLP-based multi-stage ranking formulation. Our model employs distance-based matching for candidate generation and a Transformer architecture for final matching.
Yoonhwa Jung, Jaehoon Jung, Byungil Kim, SangUk Han
Journal of Cleaner Production, 2020
Journal paper
An LSTM-RNN-based forecasting model is presented for investigation of PV sites. Time series data of spatial and meteorological conditions are considered and this work allows to search and evaluate suitable locations for PV plants in a wide area.
*Cited over 130
Peer-reviewed Conference Papers
Yoonhwa Jung, Mani Golparvar-Fard
Computing in Civil Engineering, 2024
Conference paper
This paper presents a Voice-activated Artificial Intelligence (AI) assistant, CONSTRUCTVOICEBOT. Our solution leverages the first domain-specific Speech-to-Text Transformer model, called CON-WHISPER, with a synthetic voice-text dataset from 35 commercial building projects. We delve into practical applications through use cases such as Time and Material reporting, daily construction reporting, quality assurance, and curation of construction workflows.
Juan D. Núñez-Morales†, Yoonhwa Jung†, Mani Golparvar-Fard
Proceedings of the 41th ISARC, 2024
Conference paper (†Equal contribution)
Built on UniformatBridge, ASTM Uniformat classification is utilized to map color-coded 3D point clouds aligned with schedule activities without relying on BIM as a baseline. Exemplary results on tied new transformer-based models with few-shot learning are shown.
Juan D. Núñez-Morales, Yoonhwa Jung, Mani Golparvar-Fard
Construction Research Congress, 2024
Conference paper
AIConstruct system is presented to demonstrate, for the first time, how the integration of text and image can create seamless data synchronization for construction progress monitoring and automated schedule generation, unlocking a new research paradigm.
Yoonhwa Jung, Fouad Amer, Mani Golparvar-Fard
Construction Research Congress, 2022
Journal paper (in preparation)/ Conference paper
Building on the predefined rules and heuristics formulated in the Defense Contract Management Agency (DCMA)’s 14 Point Schedule Quality Assessment, this paper explores the feasibility of heuristic-based and deep learning methods to assess a project schedule health from qualitative and quantitative perspectives.
Review
Paper
Yoonhwa Jung, Fouad Amer, Mani Golparvar-Fard
Proceedings of the 38th International Conference of CIB W78, Luxembourg, 2021
Conference paper
A close examination on the problems underpinning construction scheduling theory and practice such as sequencing logic and activity description by offering a systematic review on: 1) the way in which BIM-driven schedules are formalized; and 2) the challenges of tying in Building Information Modeling (BIM) with project schedules and/or BIM-driven schedule creation techniques.
Computer Science domain papers
Ikhyun Cho, Yoonhwa Jung, Julia Hockenmaier
EMNLP Findings, 2023
project page/ Paper
We present a simple, but effective method to incorporate syntactic dependency information directly into transformer-based language models (e.g. RoBERTa) for Aspect-Based Sentiment Classification (ABSC). Yet, SIR-ABSC outperforms these more complex models, yielding new state-of-the-art results on ABSC.
Ikhyun Cho, Yoonhwa Jung, Julia Hockenmaier
Preparing a submission, 2025
We introduce PTMoE-Cap, a simple yet effective approach for generating stylized image captions, leveraging the synergy of Mixture-of-Experts (MoE) and prompt learning techniques as a effective routing source.
Machine
Unlearning
Yoonhwa Jung, Ikhyun Cho, Shun-Hsiang Hsu, Julia Hockenmaier
Preparing a submission, 2025
project page/ arXiv
We leverage meticulously crafted adversarial noise to generate a parameter mask, effectively resetting certain parameters and rendering them unlearnable. A novel approach called Attack-and-Reset for Unlearning (ARU) outperforms current state-of-the-art results on two facial machine-unlearning benchmark datasets.
