Email: [email protected]

I am a PhD student at the Data Mining Lab in Seoul National University, advised by Professor U Kang.

My research centers on developing self-supervised learning methods that can learn effectively from data with label uncertainty — where labels are incomplete, noisy, or entirely missing. I design algorithms that adapt to diverse data types, including graphs, speech, and temporal sequences, aiming to make robust predictions without relying on large amounts of clean annotations.

These methods have been applied to real-world scenarios such as graph-based classification with incomplete supervision, domain-adapted speech recognition, and forecasting on temporal interaction networks.

Education

Seoul National University (Sep. 2020 - Current)

Ph.D. Candidate in the Graduate School of Artificial Intelligence
Expected Graduation: 2026.08

Sung Kyun Kwan University (Mar. 2015 - Aug. 2020)

B.S in Department of Computer Science and Engineering & Mathematic Science
GPA: 4.14 / 4.50; C.S.: 4.26 / 4.50

Publications

2025

Accurate Graph-based Multi-Positive Unlabeled Learning via Disentangled Multi-view Feature Propagation Junghun Kim, Ka Hyun Park, Hoyoung Yoon, and U Kang KDD 2025 [code | pdf | blog (Korean)]
Accurate Link Prediction for Edge-Incomplete Graphs via PU Learning Junghun Kim, Ka Hyun Park, Hoyoung Yoon, and U Kang AAAI 2025 [code | pdf | blog (Korean)] Accepted as oral presentation

2024

Domain-Aware Data Selection for Speech Classification via Meta-Reweighting Junghun Kim, Ka Hyun Park, Hoyoung Yoon, and U Kang Interspeech 2024 [code | pdf | blog (Korean)]
Fast and Accurate Domain Adaptation for Irregular Tensor Decomposition Junghun Kim, Ka Hyun Park, Hoyoung Yoon, and U Kang KDD 2024 [code | pdf | blog (Korean)]
Accurate Semi-supervised Automatic Speech Recognition via Multi-hypotheses-based Curriculum Learning Junghun Kim******,* Ka Hyun Park*, and U Kang (*equal contribution) PAKDD 2024 [code | pdf | blog (Korean)] Accepted as oral presentation