김현준 연구실

We develop intelligent systems that discover, retrieve, and integrate knowledge from heterogeneous sources, including text, images, and structured knowledge. Our research focuses on multimodal question answering and retrieval-augmented generation (RAG) to provide accurate and evidence-grounded answers to knowledge-intensive questions. We also explore multi-agent systems that collaboratively retrieve information, reason over multiple sources, and solve complex tasks. We develop personalized recommender systems that learn from user preferences, behavioral patterns, and sequential interactions. Our recent research focuses on generative recommendation, where recommendation is formulated as a generation problem, as well as sequential recommendation for modeling evolving user interests over time. We investigate LLM-based recommendation to predict future user behaviors from textual user context. We develop machine learning methods for modeling complex temporal patterns and predicting future events from sequential data. Our research focuses on financial time-series forecasting, capturing both short- and long-term dynamics in financial markets. We also study time-series anomaly detection to identify unusual patterns and unexpected events in large-scale temporal data. We develop efficient and scalable data mining algorithms for analyzing and retrieving information from large-scale datasets. Our research focuses on graph mining, high-dimensional retrieval, and pattern matching, with particular emphasis on reducing computational and memory costs. We aim to design algorithms that enable fast graph processing, similarity search, and pattern discovery while maintaining high retrieval and matching accuracy. We develop machine learning methods for predicting clinical outcomes and suppor