题目：Discovering causal structures using non-Gaussian structural equation models 报告人：Takashi Washio
Professor,The Institute of Scientific and Industrial Research,Osaka University 时间：11月1日（星期日），16：00-17：00 地点：蒙民伟楼404会议室 摘要：
Structural equation models (SEMs) are widely used to model data generating
processes or causal relations. However, its identifiability is not ensured
within the modeling framework based on Gaussianity of the data. Recently, a
non-Gaussian framework has been shown to be useful for discovering SEMs. In
this talk, under given data strictly following the model, we introduce a new
non-Gaussian estimation method guaranteed to converge to a unique model in a
fixed number of steps without any algorithmic parameters. 简历：
Takashi Washio is an associate professor in the division of Intelligent
Systems Science at the Institute of Scientific and Industrial Research of
Osaka University since 1996. Before joining the university, he had been a
visiting researcher of Nuclear reactor laboratory of Massachusetts Institute
of Technology (MIT) from 1988 to 1990 and a researcher in Mitsubishi Research
Institute from 1990 to 1996. His current main interest is the development of
the theories and the algorithms for the automated discovery of first principle
equations from numerical data and the development of the theories and the
algorithms for complete search to extract frequent patterns from graph
structured data. He received hid Bs, Ms, and PhD degree in nuclear engineering
from Tohoku University, and he has been a research affiliate of MIT since he
left MIT. He is now on the editorial board of New Generation Computing. He is
a member of AAAI, JSAI, IPSJ, JSFTS, and SICE. He received the JSAI best paper
awards from Japanese Society for Artificial Intelligence (2001) and the Best
Paper Awards of AESJ from Atomic Energy Society of Japan (1996). |
