I am currently serving as a Technical Research Personnel at Dable for military purposes. I received my M.S. and B.S. in School of Electrical Engineering from KAIST.

My research primarily focuses on improving recommender systems by incorporating information theory, social networks, and causal inference. I have published papers both in machine learning and information theory conferences (e.g., NeurIPS, ISIT, and IEEE TIT).

Research Interests

I have worked on numerous projects with dual objectives: fortifying the robustness within these models by removing various sources of data bias and enhancing the performance of machine learning models in data-scarce environments. Therefore, I am broadly interested in robust machine learning, integrated with causal inference and graph machine learning.

I addressed challenges related to data scarcity by utilizing graph-based information. I developed a computationally efficient algorithm that leverages user social similarity graphs as valuable side information to mitigate data scarcity. Additionally, I conducted a theoretical analysis to determine the minimum number of observed matrix entries necessary for perfect recovery. Remarkably, the proposed algorithm not only asymptotically achieved this information-theoretic limit but also surpassed the estimation performance of state-of-the-art algorithms.

Furthermore, I tackled the data bias issue through causal inference. To address the challenges, I proposed a novel conversion label generation methodology that takes advantage of counterfactual inference, enabling me to leverage the entire dataset and consequently address data bias and scarcity issues. Specifically, I extracted knowledge from the observed data and extended it to unobserved data to obtain conversion labels, allowing me to leverage a substantial amount of data in model training and consequently enhance prediction accuracy.