JB is currently a PhD student at KAIST AI, advised by Jinwoo Shin, and an AI resident at NAVER Cloud. He delves into the true nature of deep learning through the lens of human cognitive systems, with a particular focus on how humans acquire, process, and apply knowledge across multiple modalities. Before joining KAIST, JB led AI research for patent analysis and retrieval at Tanalysis. He obtained a Master’s Degree in Artificial Intelligence from Yonsei University, where he studied neural architectures and their foundations in formal language under the guidance of Yo-Sub Han. Prior to his journey into AI research, JB worked as a software engineer at Samsung Electronics.

Research Interests

JB’s research focuses on how cognitive principles can advance deep learning toward genuine artificial intelligence. Specifically, his research involves:

• Multimodality from a cognitive perspective
  Humans understand the entire world through multiple senses, yet current models process modalities separately. His research explores how human cognition can inform the design of unified multimodal models.

• Deep learning under the Poverty of the Stimulus
  Unlike humans, who acquire knowledge from only a small amount of experience, today’s large models rely on massive data. His research aims to bridge this gap by seeking optimal learning algorithms that enable models to understand the world from limited exposure.

• Knowledge representation in deep neural models
  Despite the vast knowledge encoded in deep models, accessing and leveraging such knowledge remains a challenge. His research investigates how knowledge can be represented, accessed, and modified within deep neural models.

Publications

• HyperCLOVA X 8B Omni. [arXiv] [model]
  HyperCLOVA X 32B Think. [arXiv] [model]
  NAVER Cloud HyperCLOVA X Team
  Technical Reports

• Scalable and Robust LLM Unlearning by Correcting Responses with Retrieved Exclusions. [arXiv]
  Junbeom Kim, Kyuyoung Kim, Jihoon Tack, Dongha Lim, Jinwoo Shin
  Preprint

• ATHENA: Mathematical Reasoning with Thought Expansion. [arXiv] [code]
  JB Kim, Hazel Kim, Joonghyuk Hahn, Yo-Sub Han.
  In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing (EMNLP 2023), 16315–16327, 2023

Experience

• Ph.D. in Artificial Intelligence, KAIST. ‘25 ~ Present
  (Advisor: Jinwoo Shin)

• AI Resident, NAVER Cloud. ‘25 ~ Present

• AI Research Lead, Tanalysis. ‘23 ~ ‘24

• M.S. in Artificial Intelligence, Yonsei University. ‘21 ~ ‘23
  (Advisor: Yo-Sub Han)

• Software Engineer, Samsung Electronics. ‘17 ~ ‘19