BEGIN:VCALENDAR
PRODID:-//eluceo/ical//2.0/EN
VERSION:2.0
CALSCALE:GREGORIAN
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/749
DTSTAMP:20230914T125936Z
SUMMARY:An Invitation to Causal Inference
DESCRIPTION:Speaker: Piyush Srivastava\n\nAbstract: \nConsider a system con
sisting of a binary "stimulus" variable X\, (e.g. whether an individual co
nsumes tobacco)\, a binary "response" variable Y\, (e.g. whether the indiv
idual develops a particular disease) and a "hidden" confounding variable U
\, (e.g. a hypothetical genetic factor that affects the individual's prope
nsity to both consume tobacco as well as to develop the disease). Give
n P\, the observed probability distribution of the pair (X\, Y)\, we ask:
what is the causal effect of X on Y? In particular\, can it be determined
solely from the knowledge of the "correlation" between X and Y? In the e
arly 1990s\, Judea Pearl proposed a formalization of the above question in
the language of directed graphical models. Put differently\, Pearl's
framework formalized what it might mean to compute the strength of a causa
l relationship (such as may be measured in a controlled experiment) given
only data about correlations among different components on the system.\n\n
It is intuitively clear that this problem is not always solvable: the exam
ple considered above is a case in point. However\, a long line of work b
y several researchers culminated in 2006 in a complete algorithmic charact
erization of graphical models in which the problem is solvable. Further\,
this characterization also included an algorithmic procedure which takes t
he observed distribution and outputs the requisite "causal" distribution i
n those cases where the problem is solvable [Huang and Valtorta\, 2006 and
Shpitser and Pearl\, 2006].\n\nThis talk will introduce directed graphica
l models and causal inference problem\, and then give an overview of the s
olution of Huang-Valtorta and Shpitser-Pearl. We will then look at some
recent progress on analyzing the 'robustness' (or 'condition number') of t
hese solutions with respect to "noise" or "imperfections" in the descripti
on of the model or in the measurement of the observed distribution. Surp
risingly\, even though causal inference is a statistical problem\, such ro
bustness questions were only asked (and partly answered) recently\, and se
veral exciting future directions remain open. Time permitting\, we will
then look at other related notions of causal inference\, such as the notio
n of directed mutual information in information theory\, and the theory of
linear structural equations.\n\nDisclaimer: Part of this is joint work wi
th Leonard J. Schulman (available from http://www.tifr.res.in/~piyush.sriv
astava/research.html#causal-inference). Some other parts are served with g
entle dollops of caveats emptor.\n \n
URL:https://www.tcs.tifr.res.in/web/events/749
DTSTART;TZID=Asia/Kolkata:20170127T160000
DTEND;TZID=Asia/Kolkata:20170127T173000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
END:VCALENDAR