NAVER Fortifies Global AI Leadership with Hyperscale Language Model at World-Renowned Conference
NAVER Fortifies Global AI Leadership with Hyperscale Language Model at World-Renowned Conference
NAVER Fortifies Global AI Leadership with Hyperscale Language Model at World-Renowned Conference
- NAVER to present seven papers, including key research findings on HyperCLOVA at EMNLP 2021, later this year
- NAVER presented nine papers at INTERSPEECH 2021, the highest number of presentations among Korean companies participating in the conference
- 58 papers accepted at the global top conferences as of September, showing NAVER’s significant investment in AI R&D
- NAVER to continue expanding its global AI R&D ecosystem
2021-09-03
NAVER Corporation reaffirmed its AI leadership with seven R&D papers including a study on HyperCLOVA accepted by Empirical Methods in Natural Language Processing (EMNLP) 2021. The company also successfully presented nine papers at another recent international academic conference INTERSPEECH 2021.
By doing so, the company improved on its own record, with 58 papers presented at the main sessions in top-tier AI conferences, including EMNLP, International Conference on Learning Representations (ICLR), Conference on Computer Vision and Pattern Recognition (CVPR), Association for Computational Linguistics (ACL) and many more in 2021—a significant increase over the 43 papers accepted in 2020. The increase reflects NAVER’s massive investment in AI research as it expands its global AI R&D ecosystem with leading researchers around the world.
NAVER’s papers on hyperscale AI to be presented at EMNLP reflect its R&D leadership
NAVER’s key research paper on its hyperscale AI platform, HyperCLOVA, will be the main presentation at EMNLP, which is considered the most prestigious conference on natural language processing (NLP) in AI studies. NAVER will present a total of seven papers at the conference. EMNLP will be held in the Dominican Republic from Nov. 7 to 11 this year and streamed online in real-time.
HyperCLOVA is the world’s first and biggest hyperscale AI model based on the Korean language. The language model has studied 6,500 times more Korean language data than OpenAI’s GPT-3. The paper that NAVER will present at EMNLP introduces HyperCLOVA and the data set used for training AI as well as the performance verifications of various sized AI models. The paper’s 37 authors proved that HyperCLOVA is capable of executing in-context, language learning performances on tasks in Korean. In the paper, the team also explains innovative changes in materializing the No Code AI paradigm by providing AI-prototyping capabilities to non-experts of machine learning by introducing HyperCLOVA studio, an interactive prompt engineering interface.
In addition, NAVER will also present another HyperCLOVA-related paper on a novel data augmentation technique that leverages large-scale language models to generate text samples. Other papers to be presented at the conference cover various topics, including more cost-effective extraction of information from document images and the exploration of possibly using AI language models as knowledge bases in the biomedical domain.
“NAVER has focused its investment in hyperscale AI technologies since the second half of 2020, and now we are seeing the results as we successfully introduce and commercialize HyperCLOVA technology in Korea,” said Jung-woo Ha, head of NAVER AI LAB. “Recognition from prestigious conferences, like EMNLP, acknowledges the value of Korean language-based AI beyond the natural language processing research that centers on English.”
NAVER’s leading research in speech and signal processing highlighted at INTERSPEECH
NAVER’s achievement in AI research was also highlighted at INTERSPEECH 2021, the largest AI conference focused on speech and signal processing, held from Aug. 30 to Sept. 3. At INTERSPEECH 2021, NAVER presented nine papers, and when combined with R&D papers from LINE, NAVER’s affiliate in Japan and part of its global AI R&D ecosystem, NAVER presented a total of 14 papers. The number is more than any other Korean internet company presented at the conference and the achievement is one of the highest among leading companies and research institutes in Asia.
The papers presented at INTERSPEECH 2021 covered a variety of topics related to speech and signal processing, such as voice recognition, voice synthesis, and dataset production. New technologies featured in some papers are already applied to NAVER services, contributing to an enhanced user experience. Voice synthesis quality-enhancing technology is used in various NAVER CLOVA’s voice synthesis services, such as CLOVA dubbing, CLOVA Smart Speaker, and CLOVA AiCall. NAVER's research on speaker diarization (i.e. detecting the speaker's voice based on vocal characteristics) is used to enhance its CLOVA Note service.
Among the nine papers, five were carried out in collaborative studies with leading companies and research institutes with AI R&D capability in Korea and overseas, such as EURECOM of France, Carnegie Mellon University, KAIST, Yonsei University, and LINE.
NAVER’s active investments expand its global AI R&D ecosystem
These achievements are the result of NAVER’s massive investment in AI Technology and cultivation of a global AI R&D ecosystem. Naver is investing more than 25% of its annual revenue into R&D including AI technologies and will invest tens of billions of won in hyperscale AI research over the next three years. Based on the massive investments in supercomputing infrastructure in 2020, the company developed and introduced hyperscale AI HyperCLOVA in Korea.
NAVER is expanding its domestic and global AI R&D network. Domestically, the company is cultivating the ecosystem with Korea Advanced Institute of Science and Technology (KAIST), Seoul National University, and others. Beyond Korea, NAVER is collaborating with Japan through LINE, has established joint research centers with Hanoi University of Science and Technology (HUST) and Posts and Telecommunications Institute of Technology (PTIT) in Vietnam, and signed a Memorandum of Understanding (MoU) with the University of Tübingen to establish a joint research center in July. The company is also working with NAVER LABS Europe on cooperative research.
-END-