-Great achievement at the 'Generative AI Competition Contest' hosted by the Korea Institute of Information Scientists and Engineers (Kiise) and AI company 'VAIV'

*Photo Description: A team of undergraduate and master's students from The Catholic University of Korea won the top prize at the "Creation AI Competition"

 The grand prized winner team of Generative AI Competition Contest is "Romance of the Three Kings" team of The Catholic University of Korea (President Won Jong-chul), consisting of Lee Jong-hyuk, a senior student majoring in computer science and information engineering and Won Hyun-sik, a master's student majoring in the artificial intelligence (supervised by Professor Kim Kang-min).

 This competition was hosted by AI specialist company ‘VAIV’ and the Korea Institute of Information Scientists and Engineers (Kiise) to foster an international AI ecosystem and nurture generative AI talent. A total of 18 teams participated in the preliminary round and 7 of them advanced to the final round after they competed fiercely. On the final round, participating teams presented their AI models and were judged by juries,

 The competition was conducted in a way that the participating teams used VAIV's generative AI foundation model, 'VibeGeM 2', to learn a generative AI model that summarizes and provides natural and appropriate answers when given a specific document and query, and were evaluated on this.

 The learned model was evaluated based on the generation ability evaluation (G-Eval) that considers the factuality, informativeness, and naturalness of the generated summary sentences, and the classification performance evaluation (Accuracy) that evaluates the appropriateness of the response.

 CUK team, 'Romance of the Three Kingdoms' approached the problem by considering the problem of a small dataset required for language model learning. The VibeGeM 2 model has approximately 14 billion parameters, so it was difficult to learn sufficiently with only the 50 datasets provided in the competition. To solve this, the team first used the GPT-4 API to augment the dataset to secure sufficient high-quality data.

 Afterwards, the LoRA technique was applied to enable effective learning even with a small amount of data, and training was conducted so that the model could generate natural answers through Instruction Tuning. The effectiveness of this approach was recognized and team 'Romance of the Three Kingdoms' of CUK won the grand prize.

 Data Science and Artificial Intelligence professor, Kim Kang-min said, “Through this competition, students were able to learn problem-solving skills that arise during the AI model development process,. I recommend that many students participate in this type of competition to gain experience.”