DSBA Lab

The Data Science & Business Analytics (DSBA) Lab, housed in the Department of Industrial Engineering at Seoul National University, focuses on Industrial Data Analytics, applying data science and machine learning across diverse data modalities to solve real-world industrial problems.

Time-series Analytics
Our work in time-series analytics focuses on developing advanced algorithms for anomaly detection and forecasting, along with representation learning to effectively leverage time-series data. We build self-supervised frameworks that capture key temporal and contextual patterns, design transformer-based models for detecting anomalies in multivariate data, and explore ensemble methods to enhance predictive accuracy in complex industrial settings.

Natural Language Processing
We push the boundaries of NLP by leveraging pre-trained language models and novel loss functions to solve industrial problem. Our projects include training-free log anomaly detection using token-level insights, enhancing dialogue state tracking through entity-adaptive pre-training, and building high-quality, domain-specific datasets for finance using chain-of-thought prompting. We also design robust evaluation frameworks that ensure consistency and reliability across diverse language tasks.

Computer Vision
Our computer vision research aims to develop robust and reliable image analysis systems that perform well in real-world scenarios. We focus on enhancing model accuracy and resilience, including the detection of anomalies in image data, while addressing challenges such as adversarial conditions. We propose new data augmentation techniques that maintain semantic consistency between images and text, develop student-teacher networks to bolster adversarial robustness, and employ ensemble methods for improved caption re-ranking. These efforts aim to ensure that visual recognition systems are both accurate and resilient in dynamic, real-world environments.

Applications
Beyond theoretical development, our lab emphasizes practical applications of advanced algorithms. We apply our methodologies to improve operational efficiency in semiconductor manufacturing—forecasting throughput in automated material handling systems—and to optimize maintenance strategies by accurately classifying equipment faults using shipment inspection data and maintenance reports. Our research is geared toward building intelligent, data-driven systems that enhance productivity and reliability in industrial settings.

GRL