Krafton Has 5 Papers Accepted at the World's Largest AI Conference 'NeurIPS'

Krafton announced on the 11th that it will participate in the world's largest artificial intelligence (AI) academic conference, the Conference on Neural Information Processing Systems (NeurIPS 2023, hereinafter NeurIPS), and present five papers.

NeurIPS is the most prestigious conference in the fields of AI and machine learning worldwide. It is known for its rigorous review process, with an acceptance rate of only about 26%. Krafton registered five papers in the Main Track of NeurIPS 2023, the highest number among domestic game companies. This is the third highest number even when including both domestic and international game companies.

Marking its 37th edition this year, the conference will be held from December 10 to 16 in New Orleans, USA. Krafton plans to present its research to AI scholars worldwide through the Main Track.

The topics of the papers to be presented include ▲preventing the generation of inappropriate images when using image generation AI models ▲developing reinforcement learning techniques that operate stably in procedurally generated environments and efficiently learn hierarchical task structures ▲developing small-scale language models with knowledge and reasoning abilities for specialized domains ▲developing vision and language models that perform effectively even in specialized fields ▲and robust optimization techniques that enable model training with low-quality data. These papers were highly rated as foundational technologies that can enhance productivity across game development.

Lee Kang-wook, Head of Deep Learning at Krafton, said, "Krafton is striving to secure foundational deep learning technologies and develop next-generation technical capabilities. Building on the achievements at NeurIPS, we plan to take the lead in serving as a catalyst for the expansion of the domestic game industry and AI ecosystem."

Meanwhile, this year Krafton published a total of 14 papers at international AI conferences including NeurIPS, the Association for Computational Linguistics (ACL), and the Conference on Learning Theory (COLT).