UAI paper correction
Ross D. Shachter (shachter@stanford.edu)
Thu, 30 Jul 1998 21:58:13 -07
Dear UAI colleague,
At this past week's UAI Conference, I presented a paper entitled
"Bayes-Ball: The Rational Pastime (for Determining Irrelevance and
Requisite Information)." One of the results in the paper is a sub-linear
algorithm for determining the probability distributions, deterministic
functions, and observations required to answer a given query on a belief
network.
That result, and others in the paper, builds on a seminal paper,
"Identifying Independence in Bayesian Networks," by Dan Geiger, Thomas
Verma, and Judea Pearl which appeared in Networks in 1990 (GVP90). One of
the algorithms in GVP90 determines the probability distributions required to
answer a given query, but it does not attemopt to determine which
deterministic functions and observations are required. Although GVP90 makes
clear that it is not trying to determine all of the required information, in
my UAI paper I incorrectly characterize GVP90 as having an "error" in this
algorithm. That is an unfortunate characterization, and it came about
because I did not examine GVP90 while preparing my UAI paper. Upon my
return home from UAI, I did examine GVP90 and I realized my mistake. As far
as I know, this mischaracterization of GVP90 is the only mistake in my UAI
paper. I know of no errors in GVP90 at all.
I myself have suffered from having earlier work misrepresented, and I
regret having done it to Danny, Thomas, and Judea, who I greatly respect.
Thank you for alllowing me to correct my mistake for the record,
Ross