Jang Jaeseok, a master’s student in the Big Data-Driven Artificial Intelligence Lab of the Department of Data Science at Seoul National University of Science and Technology (Seoul Tech), led by Professor Kwon Hyuk-Yoon http://bigdata.seoultech.ac.kr), has been selected to present his journal at NeurIPS 2024, the world’s largest conference on artificial intelligence, to be held in Vancouver, Canada, in December 2024. 
 
Jaeseok Jang and Hyuk-Yoon Kwon, "Are Multiple Instance Learning Algorithms Learnable for Instances?," Thirty-Eighth Annual Conference on Neural Information Processing Systems (NeurIPS 2024), Vancouver, Canada, Dec. 2024 (Full journal text: https://openreview.net/pdf?id=cUcvlgkQxP).
 

The proposed method addresses Multiple Instance Learning (MIL) to detect individual instance-level anomalies using only a small amount of bag-level labeling information. MIL is an innovative learning model that can achieve high accuracy d on correlation analysis between low-level instances with only low-cost high-level labeling. The research provides PAC-d theoretical proof of the feasibility of instance-level learning for previously proposed MIL methodologies and performs experimental validation. In particular, this study offers a theoretical verification of the limitations of the attention-d aggregation methodology commonly used in existing studies and demonstrates the effectiveness of the conjunctive-d aggregation methodology, which provides a direction for future research in this field.
 

NeurIPS is ranked in the top tier of BK-recognized international conferences and is listed among the best conferences selected by the Korean Institute of Information Scientists and Engineers. Accordingly, Jang’s journal will be presented as the only NeurIPS paper selected for the Top Conference Session, the 2024 Computer Software Conference by the same institute, to be held in Yeosu on Dec. 18 (https://www.kiise.or.kr/conference/main/getContent.do?CC=KSC&CS=2024&content_no=2147&PARENT_ID=011600).
 
* This research was supported by Seoul Tech’s BK21 Data Science, Business Potential Education Research Center, and Research Foundation.