{"@context":"https://neupai.io/schema/v0.2","@type":"StructuredNewsArticle","identity":{"article_id":"tech42_20260825_nota-emnlp-2026-moe-quantization","canonical_url":"https://www.tech42.co.kr/%ec%a0%84%eb%ac%b8%ea%b0%80-%ec%84%a0%ed%83%9d-%ed%9d%94%eb%93%a4%eb%a6%bc-%ec%a4%84%ec%9d%b8%eb%8b%a4%eb%85%b8%ed%83%80-emnlp-2026%ec%84%9c-moe-%ec%96%91%ec%9e%90%ed%99%94/?utm_source=rss&utm_medium=rss&utm_campaign=%25ec%25a0%2584%25eb%25ac%25b8%25ea%25b0%2580-%25ec%2584%25a0%25ed%2583%259d-%25ed%259d%2594%25eb%2593%25a4%25eb%25a6%25bc-%25ec%25a4%2584%25ec%259d%25b8%25eb%258b%25a4%25eb%2585%25b8%25ed%2583%2580-emnlp-2026%25ec%2584%259c-moe-%25ec%2596%2591%25ec%259e%2590%25ed%2599%2594","ai_url":null,"publisher":{"name":"테크42","domain":"www.tech42.co.kr","type":"online"},"author":"김한수 기자","published_at":"2026-08-25T00:06:54.000Z","updated_at":null,"language":"ko","article_type":"straight_news","originality":"self_produced"},"content":{"headline":"‘전문가 선택’ 흔들림 줄인다…노타, EMNLP 2026서 MoE 양자화 논문 2편 채택","summary":"AI 모델 최적화 기업 노타가 EMNLP 2026에서 초거대 AI 모델의 양자화 성능 저하 문제를 해결하는 연구 논문 2편이 채택됐다. 노타는 Qwen3.8-Max 등 초대형 모델의 GPU 사용량을 대폭 줄이는 기술력을 입증했다.","topics":["AI","머신러닝","반도체","클라우드","연구개발"],"geography":["KR"],"entities":[{"name":"노타","canonical_id":"corp:kr:nota","type":"company","role_in_article":"primary_subject","metadata":{"ticker":null,"parent":null}},{"name":"EMNLP 2026","canonical_id":"org:us:emnlp-2026","type":"organization","role_in_article":"mentioned","metadata":{"ticker":null,"parent":null}},{"name":"Qwen3.8-Max","canonical_id":"product:cn:qwen3-8-max","type":"product","role_in_article":"mentioned","metadata":{"ticker":null,"parent":"corp:cn:alibaba"}},{"name":"엔비디아","canonical_id":"corp:us:nvidia","type":"company","role_in_article":"mentioned","metadata":{"ticker":"NVDA","parent":null}},{"name":"문샷AI","canonical_id":"corp:cn:moonshot-ai","type":"company","role_in_article":"mentioned","metadata":{"ticker":null,"parent":null}},{"name":"Kimi K3","canonical_id":"product:cn:kimi-k3","type":"product","role_in_article":"mentioned","metadata":{"ticker":null,"parent":"corp:cn:moonshot-ai"}},{"name":"업스테이지","canonical_id":"corp:kr:upstage","type":"company","role_in_article":"mentioned","metadata":{"ticker":"038340.KQ","parent":null}},{"name":"Solar Open 2","canonical_id":"product:kr:solar-open-2","type":"product","role_in_article":"mentioned","metadata":{"ticker":null,"parent":"corp:kr:upstage"}},{"name":"채명수","canonical_id":"person:kr:chae-myeong-soo","type":"person","role_in_article":"quoted","metadata":{"ticker":null,"parent":"corp:kr:nota"}},{"name":"ICML 2026","canonical_id":"org:us:icml-2026","type":"organization","role_in_article":"mentioned","metadata":{"ticker":null,"parent":null}},{"name":"Qwen3.5-4B","canonical_id":"product:cn:qwen3-5-4b","type":"product","role_in_article":"mentioned","metadata":{"ticker":null,"parent":"corp:cn:alibaba"}}],"claims":[{"id":"c1","statement":"노타의 초거대 AI 모델 최적화 연구 논문 2편이 EMNLP 2026 메인 컨퍼런스와 파인딩스에 각각 채택됐다","as_of":"2026-08","as_of_explicit":false,"as_of_raw":"25일","source_type":"company_disclosure","comparison":null,"type":"fact","figures":{"value":2,"unit":"count","approximate":false,"converted":null},"expiry_hint":null,"insight":null},{"id":"c2","statement":"EMNLP 2026 대회에는 약 1만8000편의 논문이 제출됐다","as_of":"2026","as_of_explicit":false,"as_of_raw":"올해","source_type":"company_disclosure","comparison":null,"type":"fact","figures":{"value":18000,"unit":"count","approximate":true,"converted":null},"expiry_hint":null,"insight":null},{"id":"c3","statement":"EMNLP 2026 메인 컨퍼런스 채택률은 15.4%였다","as_of":"2026","as_of_explicit":false,"as_of_raw":"올해","source_type":"company_disclosure","comparison":null,"type":"fact","figures":{"value":15.4,"unit":"%","approximate":false,"converted":null},"expiry_hint":null,"insight":null},{"id":"c4","statement":"노타는 3종의 MoE 모델에 4비트와 3비트 양자화를 적용해 실험했다","as_of":"2026-08","as_of_explicit":false,"as_of_raw":"2026년 8월","source_type":"company_disclosure","comparison":null,"type":"fact","figures":{"value":3,"unit":"count","approximate":false,"converted":null},"expiry_hint":null,"insight":null},{"id":"c5","statement":"노타는 ICML 2026 'Efficient Qwen Competition'에서 약 40개 참가팀 가운데 3위를 차지했다","as_of":"2026-07","as_of_explicit":false,"as_of_raw":"지난달","source_type":"company_disclosure","comparison":null,"type":"fact","figures":{"value":3,"unit":"rank","approximate":false,"converted":null},"expiry_hint":null,"insight":null},{"id":"c6","statement":"경진대회에서 오픈소스 LLM 'Qwen3.5-4B'를 엔비디아 A10G GPU 한 장에서 구동했다","as_of":"2026-07","as_of_explicit":false,"as_of_raw":"지난달","source_type":"company_disclosure","comparison":null,"type":"fact","figures":{"value":1,"unit":"count","approximate":false,"converted":null},"expiry_hint":null,"insight":null},{"id":"c7","statement":"노타는 모델 성능을 유지하면서 기존보다 평균 6.978배 빠른 추론 속도를 구현했다","as_of":"2026-07","as_of_explicit":false,"as_of_raw":"지난달","source_type":"company_disclosure","comparison":"previous_version","type":"fact","figures":{"value":6.978,"unit":"ratio","approximate":false,"converted":null},"expiry_hint":null,"insight":null},{"id":"c8","statement":"1조개가 넘는 매개변수를 보유한 'Qwen3.8-Max'는 구동에 필요한 엔비디아 B300 GPU 수가 24장에서 4장으로 줄었다","as_of":"2026-08","as_of_explicit":false,"as_of_raw":"2026년 8월","source_type":"company_disclosure","comparison":"before_optimization","type":"fact","figures":{"value":4,"unit":"count","approximate":false,"converted":null},"expiry_hint":null,"insight":null},{"id":"c9","statement":"문샷AI의 'Kimi K3'는 B300 GPU 8장에서 최대 4장으로 축소했다","as_of":"2026-08","as_of_explicit":false,"as_of_raw":"2026년 8월","source_type":"company_disclosure","comparison":"before_optimization","type":"fact","figures":{"value":4,"unit":"count","approximate":false,"converted":null},"expiry_hint":null,"insight":null},{"id":"c10","statement":"업스테이지의 'Solar Open 2'는 H100 GPU 8장에서 2장으로 축소했다","as_of":"2026-08","as_of_explicit":false,"as_of_raw":"2026년 8월","source_type":"company_disclosure","comparison":"before_optimization","type":"fact","figures":{"value":2,"unit":"count","approximate":false,"converted":null},"expiry_hint":null,"insight":null},{"id":"c11","statement":"노타는 모바일 엣지 AI, 생성형 AI, 비전언어모델(VLM), LLM 분야에서 50여건의 연구 결과를 국내외 학회와 학술지에 발표했다","as_of":"2026-08","as_of_explicit":false,"as_of_raw":"그동안","source_type":"company_disclosure","comparison":null,"type":"fact","figures":{"value":50,"unit":"count","approximate":true,"converted":null},"expiry_hint":null,"insight":null},{"id":"c12","statement":"AI 모델이 수천억, 수조개의 매개변수 규모로 커지면서 산업의 경쟁축이 모델 자체의 성능을 넘어 운영 효율로 이동하고 있다","as_of":"2026-08","as_of_explicit":false,"as_of_raw":"2026년 8월","source_type":"company_disclosure","comparison":null,"type":"opinion","figures":null,"expiry_hint":null,"insight":{"analyst_name":"채명수","analyst_affiliation":"노타","confidence":"high","forecast_horizon":"long_term","sentiment":"neutral","reasoning":"모델 규모 확대로 인해 운영 효율이 경쟁력의 핵심 축으로 부상하고 있기 때문"}},{"id":"c13","statement":"최신 AI 모델의 구조 변화에 선제적으로 대응할 수 있도록 연구개발 투자와 지원을 이어갈 것","as_of":"2026-08","as_of_explicit":false,"as_of_raw":"2026년 8월","source_type":"company_plan","comparison":null,"type":"future_plan","figures":null,"expiry_hint":"2026-12","insight":null}],"ai_emotional_context":{"valence":0.4,"arousal":0.3,"primary_emotions":[{"emotion":"optimistic","intensity":0.5}],"secondary_emotions":[{"emotion":"calm","intensity":0.3}],"emotional_triggers":[{"claim_id":"c8","emotion":"optimistic","reason":"significant_positive_metric"}]},"image":{"url":"https://onjblseywainslkhvkav.supabase.co/storage/v1/object/public/article-images/1e47966f-7e89-4534-8167-591e38201e97.jpg","alt":"노타 로고와 EMNLP2026 부다페스트 표기 아래 MoE 양자화 관련 논문 두 편의 제목이 담긴 카드 이미지.","caption":"노타는 ‘EMNLP 2026’에서 초거대 AI 모델 최적화 연구 논문 2편이 각각 메인 컨퍼런스와 파인딩스(Findings)에 채택됐다고 25일 밝혔다.","source":"first_img","alt_status":"auto"}},"provenance":{"source_chain":["primary_reporting"],"original_source_url":null,"related_articles":[]},"temporal":{"freshness":"recent","next_update_expected":null},"access":{"license":"neupai_standard","attribution_required":true,"structured_data":"free","full_text_available":false,"full_text_access":null}}