Jeonbuk National University (JBNU) Department of Software Engineering AI&SE Lab. (AI&SE Lab., Advisor: Professor Deok-san Ryu) undergraduate researcher Eun-jin Shim recently presented the paper "Software Defect Prediction Using an In-Context Learning-Based Tabular Foundation Model (TabICL)" at the 2026 Korean Software Engineering Conference (KCSE 2026) and received the Best Short Paper Award.

This paper is a study that analyzed the applicability of TabICL, a tabular foundation model based on In-Context Learning that has recently attracted attention, to the Software Defect Prediction (SDP) problem.

Unlike conventional supervised learning methods that require large-scale training data and repeated retraining, this study proposed a new approach that performs defect prediction at the inference stage using only a few examples, without separate model training.

The research team compared and analyzed performance against existing machine-learning-based defect prediction models using publicly available software defect datasets. The results confirmed the potential for practical application while minimizing training costs. In particular, it was evaluated as having high academic and practical value because it can be applied even in environments with limited large-scale training infrastructure.

At the conference, researchers from the AI&SE Lab led by Professor Deok-san Ryu presented a total of eight papers, including Eun-jin Shim's award-winning paper, showcasing active research outcomes.

Eun-jin Shim stated, "Although this was my first conference paper, I was able to achieve meaningful results thanks to my advisor's guidance and the help of lab members. I hope to use this experience to further pursue research in software engineering and artificial intelligence."