ETRI Concludes 5th Human Understanding AI Paper Competition…409 Teams, 864 Participants
Published2026.10.19 12:35
대회는 과학기술정보통신부와 국가과학기술연구회(NST)의 후원으로 지난 4월부터 6개월간 진행됐으며, 데이콘(DACON)이 운영을 담당했다.
총 409개 팀, 864명이 참가했으며 이 중 22개 팀이 국제학술대회 ‘ICTC 2026’의 별도 트랙에서 연구 논문을 발표했다.
참가자들은 일상에서 수집되는 활동·신체 상태 데이터를 바탕으로 수면의 질, 잠들기 전 피로도와 스트레스, 총 수면시간, 수면 효율, 수면 중 각성 시간 등 7개 지표를 예측하는 AI 모델을 개발했다.
참가팀은 대회 사이트에 예측 결과를 제출해 모델 성능을 확인한 뒤 연구 내용을 논문으로 정리해 ICTC 2026에 제출했으며, 심사를 통과한 팀은 코드와 모델 설명서를 추가로 제출해 결과 재현 여부를 검증받았다.
대상은 고려대학교 ‘멀왕’ 팀(김규빈·박찬호·이규호·김은태)이 차지했으며 과학기술정보통신부장관상을 받았다.
이 팀은 개인의 평소 상태와 최근 생활 기록을 함께 분석해 예측 정확도를 높이고, 평소와 다른 변화가 감지될 때 예측값을 조정하는 알고리즘을 제안했다.
최우수상은 충남대학교 ‘LOV3’ 팀(정준모·김윤재·김민수)이 받았으며, 다차원 앙상블 스택(Multi-Axis Ensemble Stack) 구조를 활용해 개인별 차이와 최근 상태 변화를 종합 반영하는 방법을 제시했다.
우수상은 충남대학교 ‘9월 입대’ 팀, 장려상은 고려대학교 ‘춘자’ 팀과 Begas·경희대학교·Nexon 연합 ‘대자연’ 팀에 돌아갔다.
최우수상 이하 수상팀에는 한국전자통신연구원장상과 상금이 수여됐다.
ETRI 신형철 휴먼증강연구본부장은 “이번 대회는 예측 성능뿐 아니라 연구 결과의 재현성과 참신한 아이디어까지 종합적으로 평가해 우수 AI 연구를 발굴했다는 의미가 있다”며 “앞으로도 연구 데이터 공개와 국내외 연구진과의 협력을 확대해 나가겠다”고 밝혔다.
Comments
Related news
ETRI Achieves 1st and 2nd Place in Two Divisions of International AI Traffic Video Analysis Competition
The Electronics and Telecommunications Research Institute (ETRI), in collaboration with the University of Washington in the United States, participated in the ECCV 2026 AI City Challenge and achieved 1st place in the PSI-VQA division for pedestrian crossing intent analysis and 2nd place in the FETV division for intersection traffic violation analysis. By combining traffic video analysis technology accumulated over 30 years with the latest vision-language model (VLM), the integrated traffic video AI 'UniTraffic' processes diverse camera footage through a single system and enhances judgment reliability.
2026.09.22by 배종인 기자
ETRI Wins IDEA Design Award with Medical Humanoid Interface 'NAMI.I'
The user interface NAMI.I for the medical auxiliary humanoid robot NAMI, developed by the Electronics and Telecommunications Research Institute (ETRI), has won the main award at the American IDEA Design Awards. The user-centered interface was recognized for its design that enables medical staff to intuitively instruct tasks without requiring specialized robot knowledge. ETRI is promoting actual medical field demonstrations at Keimyung University Dongsan Hospital and plans to advance human-robot interaction technology based on collected feedback.
2026.09.29by 배종인 기자
ETRI and 9 Government-Funded Research Institutes Hold 'OST 2026 Open Source Tech Day'
ETRI and 9 government-funded research institutes will hold the 'OST 2026 Open Source Tech Day' at the aT Center in Seoul on the 7th. Under the theme of 'Open Source Innovation, Creating the Future Together with AI,' this event is a forum to discuss open source technology and R&D collaboration directions in the AI era. A total of 18 specialized presentations on agentic AI, AI safety and reliability, and digital sovereignty will be conducted across 3 tracks, with participation from industry, academia, and developer communities to expand collaboration between government-funded research institutes and industry.
2026.10.06by 배종인 기자
ETRI Unveils 3 Next-Generation Media Technologies Including AI-Dedicated Video Compression
ETRI has unveiled three next-generation media technologies optimized for AI. These include AI-based video encoding (VCM), 3D human generation based on sparse cameras, and generative AI-based pre-visualization technology. In environments where AI directly processes video such as autonomous vehicles, robots, and drones, these technologies enable efficient compression and transmission of large-scale data, and can be applied across various industrial sectors including smart cities, video security, and sports broadcasting.
2026.10.12by 배종인 기자
ETRI Develops LLM Technology for Automatic Configuration of Optical Access Networks in Natural Language
ETRI has developed technology to automatically configure and control optical access networks using large language models (LLM). When users issue commands in natural language, AI analyzes the intent and controls equipment. The organization has also secured pre-verification technology that validates operations in a virtual environment before actual deployment, as well as local LLM technology for security. Natural language-based network operations are now possible without complex command syntax, enabling the realization of communication network management automation.
2026.10.14by 배종인 기자
