닫기
Loading..

전자정보연구정보센터 ICT 융합 전문연구정보의 집대성

홈 홈 > 연구문헌 >

빠른 프로토타입 검출을 통한 그래프 신경망 설명

Current Result Document : 0 / 0

한글제목(Korean Title) 빠른 프로토타입 검출을 통한 그래프 신경망 설명
영문제목(English Title) Turbo-PAGE: GNN Explanations via Fast Prototype Discovery
저자(Author) Yong-Min Shin   Sun-Woo Kim   Won-Yong Shin  
원문수록처(Citation) VOL 77 NO. 01 PP. 0096 ~ 0096 (2022. 02)
한글내용
(Korean Abstract)
영문내용
(English Abstract)
Graph neural networks (GNNs) has shown great successes in graph mining tasks. However, explaining the decisions from GNN models is largely underexplored in the literature since neural networks are black boxes. Improving upon our earlier work on GNN explanations, termed prototype-based GNN explanations (PAGE), we present Turbo-PAGE, which discovers the common subgraph pattern more efficiently by handling multiple graphs at once. Experiments using synthetic and real-world datasets demonstrate that Turbo-PAGE produces better explanations compared to the state-of-the-art (SOTA) method.
키워드(Keyword)
원문 PDF 다운로드