Naver Corporation Has 10 Papers Accepted at NeurIPS 2025, Demonstrating Global Technological Prowess

[Financial News] Team Naver achieved a significant milestone with 10 papers accepted at the world-renowned Conference on Neural Information Processing Systems (NeurIPS) 2025. In addition to showcasing its research capabilities, Team Naver introduced key technologies that enhance the competitiveness of its AI services, including HyperCLOVA X, reaffirming its leadership in AI.
According to Naver Corporation on the 9th, Team Naver shared its unique AI full-stack strategy with top global research talent at this year’s conference under the theme 'From Research to Reality.' The strategy spans from research to its proprietary large-scale AI model, HyperCLOVA X, on-service AI, and industry-specific vertical services.
The ten papers accepted this year focus on practical applications of AI full-stack technologies for efficient and safe deployment in real-world services and industries. The research highlights three major areas of innovation: efficient operation of large-scale AI, enhanced safety of generative AI, and expansion into the physical world and robotics.
Key technologies for reducing service costs and improving the speed of Large Language Models (LLM) include a cache compression technology that selectively compresses important data during long-context processing, reducing memory usage by up to four times and doubling speed without sacrificing performance. Another study demonstrated that removing unnecessary intermediate steps in the Chain of Thought (CoT) process, which LLMs generate when solving complex problems, maintains accuracy while improving computational efficiency.
Research aimed at improving the quality and safety of AI-generated outputs also drew attention. This includes a new benchmark for precisely evaluating content quality in conversational search environments, a text embedding technology that more accurately reflects user intent during image generation, and a safety control technology that suppresses harmful content generation without additional retraining.
Studies expanding AI into the real world were also introduced. Notable papers presented methods for compressing scene information into bottleneck tokens to improve learning efficiency in environments where temporal continuity is critical, such as robotics and video, and proposed model architectures that help robots remember changing spaces and autonomously find paths.
Naver Corporation also actively built networks with global talent at this year’s NeurIPS. During the conference, which brought together over 29,000 AI researchers from around the world, the company operated an integrated booth to showcase its major research achievements and AI technologies.
A NAVER Cloud representative stated, “Having 10 papers accepted at NeurIPS, where the acceptance rate is only about 25%, is evidence that Naver Corporation’s research capabilities and technological prowess are recognized globally. We will continue to expand the global AI ecosystem and further strengthen our research capabilities so that AI can be implemented in real-world services and industries.”
Meanwhile, NeurIPS (Conference on Neural Information Processing Systems) is the world’s most prestigious conference on artificial intelligence and machine learning, established in 1987. Now in its 39th year, NeurIPS was held in San Diego, California, from December 2 to 7.


yjjoe@fnnews.com Jo Yoon-joo Reporter