On-Going Projects
DB4DL: High-Usability and Performance In-Memory Distributed DBMS for Deep Learning
Principle Investigator | Institute for Information & Communications Technology Planning & Evaluation (IITP) | 4/1/2020 ~ 12/31/2027
We are developing an in-memory distributed DBMS in support of deep learning, focusing on data preparation (preprocessing) technologies in the first stage and distributed deep-learning technologies in the second stage. See the project homepage for details.
Robust, Fair, Extensible Data-Centric Continual Learning
Principle Investigator | Institute for Information & Communications Technology Planning & Evaluation (IITP) | 4/1/2022 ~ 12/31/2026
We are developing continual learning (lifelong learning) methodologies which are robust to data noise, fair to data bias, and extensible to graph data, in support of data-centric AI. This project is conducted by four KAIST professors and SIA.
Core Technologies for Crowd Management Systems Using AI and Mobility Big Data
Principle Investigator | National Research Foundation of Korea (NRF) | 3/1/2023 ~ 2/28/2026
We are developing a crowd management system, which estimates the PoI-level locations of the people from their cellular trajectories and finds highly-populated regions from the population density stream in real time. This project is expected to prevent human stampedes.
Online Deep Learning Technology for Anomaly Detection from Sensor Stream Big Data
Principle Investigator | Samsung Electronics Co., Ltd. | 9/16/2020 ~ 9/15/2025
We are developing online deep learning techniques for detecting anomalies in sensor stream big data. Our techniques are designed to quickly adapt to various concept drifts and can be further improved by fine tuning and active learning.
Improving the Responsibility of LLMs via Unlearning on Intel Gaud
Principle Investigator | Naver Cloud (and Intel) | 7/1/2024 ~ 6/30/2027
We are developing novel LLM unlearning algorithms which can run on Intel Gaudi2 AI Accelerator. These algorithms are designed to effectively address the over-forgetting problem. This project is supported by NAVER-Intel-KAIST Joint AI Research Center.
Data Augmentation Technology using Synthesized Data for Improving AI Model Performance
Principle Investigator | Samsung Electronics Co., Ltd. | 5/31/2024 ~ 5/30/2025
We are attempting to mitigate various imbalances, such as class imbalance, scale imbalance, and spatial imbalance, in the training data for object detection thorough data augmentation using synthesized images. This research aims to exploit the recent advance of generative AI (e.g., Stable Diffusion) for addressing the low quality of the training data.
Previous Projects
Development of Real-Time Service Incident Prediction Technology
Principle Investigator | Samsung Electronics Co., Ltd. | 7/1/2021 ~ 7/31/2023
We developed deep learning techniques for predicting service incidents in real time through timeseries anomaly detection. Our techniques can incorporate diverse contextual information to recognize the real-world situation.
AutoML for Deep Learning Analytics of Multivariate Stream Big Data
Principle Investigator | National Research Foundation of Korea (NRF) | 9/1/2020 ~ 2/28/2023
We developed a Stream AutoML framework for clustering, classification, regression, and anomaly detection tasks so as to promote deep learning analytics of multivariate stream big data. This is a follow-up study of the NRF strategic project.
Evolutionary Self-Learning for Multivariate Stream Big Data in Smart Factories
Principle Investigator | National Research Foundation of Korea (NRF) | 11/1/2017 ~ 10/31/2020
We have developed a framework of systematically mining analysis purpose (or task) in smart factory data in order to narrow the data gap between the data and new technologies (e.g., deep learning) so as to greatly improve the usability of the data.
Geospatial Big Data Management, Analysis and Service Platform Technology Development
Principle Investigator | Korea Agency for Infrastructure Technology Advancement / Ministry of Land, Infrastructure and Transport | 8/14/2014 ~ 12/31/2019
We have developed an interactive data analytics platform for real-time geospatial big data. The platform consists of two main components: spatial complex event processing (CEP) and spatial online analytical processing (OLAP).
Machine Learning Technology for Real-Time Anomaly Detection in Semiconductor Processes
Principle Investigator | Samsung Electronics Co., Ltd. | 7/1/2018 ~ 6/30/2020
We have developed machine learning techniques for detecting outliers in real time from multivariate timeseries, which are typically generated from semiconductor manufacturing equipment, aiming at significantly improving both latency and throughput.
Discovering Periodic Human Behavior from Multivariate Sensor Data
Principle Investigator | Office of Naval Research Global (ONRG) | 1/5/2018 ~ 1/4/2020
In order to understand routine human behavior more naturally and precisely from multivariate sensor data, we have attempted to define a new form of periodic patterns and develop an efficient and light-weighted algorithm of finding these patterns.
Study on Detection of Geo-Social Patterns from Big Data Social Networks
Principle Investigator | National Research Foundation of Korea (NRF) | 5/1/2015 ~ 4/30/2018
We have designed the algorithms of finding geo-social patterns from geo-social networks and developed distributed and parallel algorithms on a memory-resident distributed platform. This project is the follow-up support of an outstanding New Faculty project by the NRF.
Prediction of Customer Revisit Intention using Indoor Movements in Stores
Principle Investigator | Microsoft Research Asia | 2/15/2017 ~ 2/14/2018
We have investigated the factors influencing customer revisit intention to offline stores by analyzing indoor movements and developed the methodology of constructing a predictive model. For this study, we obtained a license for the WIFI router data sets that hold the information of indoor movements in seven stores.
Development of Device Collaborative Giga-Level Smart Cloudlet Technology
Co-Investigator | Institute for Information & Communications Technology Promotion | 4/1/2013 ~ 2/28/2018
We have extended data mining algorithms such that they can operate on a cluster of mobile devices, which we call a smart cloudlet, and defined the API for smart cloudlet applications.
WiseKB: Big Data Based Self-Evolving Knowledge Base and Reasoning Platform
Co-Investigator | Saltlux Inc. / Institute for Information & Communications Technology Promotion | 5/1/2013 ~ 2/28/2017
We have studied the construction of knowledge bases that can self-evolve through newly learned knowledge from documents, working toward the goal of building a system like IBM Watson.
Study on Community Detection Algorithms for Big Data Social Networks
Principle Investigator | National Research Foundation of Korea (NRF) | 5/1/2012 ~ 4/30/2015
We have scaled out community detection algorithms using distributed parallel computing platforms such as Hadoop and GraphLab.
Research on Base Technologies for Development of Next-Generation Mobile-Commerce Service Platforms
Principle Investigator | Samsung Electronics Co., Ltd. | 7/3/2012 ~ 8/31/2014
We have developed a location-based social search engine using geo-social networking services such as Foursquare and developed a novel approach to extracting user interests from social curation services such as Pinterest.
Research and Development of an Educational Material Search System for the Next-Generation Education Environment
Co-Investigator | National Research Foundation of Korea (NRF) | 9/1/2011 ~ 8/31/2014
We have developed a search engine CourseShare for educational materials (e.g., slides). My team extracted multiple features from educational materials to build a multi-layer network and developed community detection algorithms for such networks.
Value Investment Analysis Chart Advancement and Convergence Platform Development
Co-Investigator | CS co., Ltd. / Small and Medium Business Administration | 6/1/2013 ~ 5/31/2015
We have developed a new methodology of predicting stock purchase timing based on the financial information of companies.
Development of Accurate and Trustable Recommender Systems
Principle Investigator | KAIST (new faculty start-up fund) | 7/1/2010 ~ 12/31/2013
We have studied pattern discovery and destination prediction from GPS trajectories in support of accurate location recommendation.