BEGIN:VCALENDAR
VERSION:2.0
PRODID:researchseminars.org
CALSCALE:GREGORIAN
X-WR-CALNAME:researchseminars.org
BEGIN:VEVENT
SUMMARY:Elina Robeva (University of British Columbia)
DTSTART:20210415T163000Z
DTEND:20210415T173000Z
DTSTAMP:20260422T054456Z
UID:SFUQNTAG/23
DESCRIPTION:Title: <a href="https://researchseminars.org/talk/SFUQNTAG/23/
 ">Hidden Variables in Linear Causal Models</a>\nby Elina Robeva (Universit
 y of British Columbia) as part of SFU NT-AG seminar\n\n\nAbstract\nIdentif
 ying causal relationships between random variables from observational data
  is an important hard problem in many areas of data science. The presence 
 of hidden variables\, though quite realistic\, pauses a variety of further
  problems. Linear structural equation models\, which express each variable
  as a linear combination of all of its parent variables\, have long been u
 sed for learning causal structure from observational data. Surprisingly\, 
 when the variables in a linear structural equation model are non-Gaussian 
 the full causal structure can be learned without interventions\, while in 
 the Gaussian case one can only learn the underlying graph up to a Markov e
 quivalence class. In this talk\, we first discuss how one can use high-ord
 er cumulant information to learn the structure of a linear non-Gaussian st
 ructural equation model with hidden variables. While prior work posits tha
 t each hidden variable is the common cause of two observed variables\, we 
 allow each hidden variable to be the common cause of multiple observed var
 iables. Next\, we discuss hidden variable Gaussian causal models and the d
 ifficulties that arise with learning those. We show it is hard to even des
 cribe the Markov equivalence classes in this case\, and we give a semi alg
 ebraic description of a large class of these models.\n
LOCATION:https://researchseminars.org/talk/SFUQNTAG/23/
END:VEVENT
END:VCALENDAR