Скачать с ютуб [Paper Review] Unified Training of Universal Time Series Forecasting Transformers в хорошем качестве

[Paper Review] Unified Training of Universal Time Series Forecasting Transformers

발표자: 석박통합과정 강형원 1. 논문 제목: Unified Training of Universal Time Series Forecasting Transformers (Woo, Gerald, et al, arXiv preprint 2024) 링크: https://arxiv.org/abs/2402.02592 2. 논문 Overview Universal forecasting을 위한 새로운 transformer architecture(MOIRAI) 제안 Large Time Model의 사전 학습을 강화하기 위한 새로운 대규모 개방형 시계열 데이터 셋 컬렉션인 LOTSA를 공개 Full-shot model과 비교했을 때 zero-shot forecaster로서 경쟁력이 있거나 우수한 성능을 달성 3. keyword: Time series, Forecasting, Probabilistic forecasting, Foundation model, Large Time-series Model, LOTSA, Rotary, Position Embedding, Distribution