Rich Ulrich <[EMAIL PROTECTED]> wrote:

> On Mon, 20 Oct 2003 18:45:03 -0700, "albinali" <[EMAIL PROTECTED]>
> wrote:
> 
> >              .... The data set set collected from the space has
> > sensor readings along with the activities. Moreover, I want to 
> > construct a bayesian network for every activity. 
> 
> Is this something I should know, 
> 'What's a Bayesian network'?

Given a joint distribution over several variables, in principle
one can always compute the marginal distribution over some of them,
perhaps conditional on known values of some others, by integrating
the joint distribution.

A joint distribution specified as a product of conditional
distributions can be given a pictorial form by drawing an oval
for each variable and drawing arrows from the variables on the
right side of the bar to the variables on the left side of the 
bar. E.g., p(A,B,C) = p(C|B) p(B|A) p(A) yields (A) -> (B) -> (C).
There are other ways to decompose a joint distribution which
lead to an undirected graph or a mixed directed-undirected graph.

It turns out that conditional independence can be determined
by studying the graph (ovals and arrows). This is important 
because conditional independence simplifies the calculations;
a problem which is intractable may become tractable if we can
find a simpler equivalent form.

What is "Bayesian" about this scheme is that the variables
A, B, C, what have you, can be any kind of quantities; 
experimental outcomes, parameters, hypotheses, etc. -- not
restricted to "random variables". A Bayesian network is just
a graphical representation of an ordinary Bayesian model.

See, for example, http://auai.org/ and the links there.

For what it's worth,
Robert Dodier
--
If I have not seen as far as others, it is because 
giants were standing on my shoulders. -- Hal Abelson
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to