The following papers are now available on the web.
Comments are highly appreciated.
Enjoy,
-----Rina.
R. Dechter, "Bucket elimination: a unifying framework for
structure-driven inference". An ICS technical report, October 1998.
<ftp://ftp.ics.uci.edu/pub/CSP-repository/papers/bucketelimR48b.ps>
The paper extends related papers
(one appearing in UAI96 and one in "Learning and inference in graphical
models" edited by Mike Jordan).
The current version provides a more complete overview of
the framework for deterministic networks,
provides a complete treatment
for probabilistic networks and optimization tasks,
and expands on the relationship
between bucket-elimination on one hand and the poly-tree and
join-tree clustering on the other.
Hybrids with conditioning search are also presented.
R. Dechter and I. Rish "Mini-buckets; a general scheme
for approximating inference", ICS Technical report, October 1998.
<ftp://ftp.ics.uci.edu/pub/CSP-repository/papers/mini-bucketsR63a.ps>
The paper expand on two earlier papers appearing in UAI97 and UAI98.
It describes a class of algorithms approximating
the bucket-elimination scheme, and thus are applicable
to constraint satisfaction, combinatorial optimization and
probabilistic inference. The approximation algorithms
are resource adjustable. The paper provides an
empirical evaluation on randomly generated networks and on
real-life domains such as medical diagnosis and probabilistic decoding.
I. Rish and R. Dechter "On the impact of causal independence"
An ICS-UCI technical report, October 1998.
<ftp://ftp.ics.uci.edu/pub/CSP-repository/papers/impactR67.ps>
The paper studies the performance benefits of
probabilistic networks having causally independent relationships.
In particular, it ties the performance of
algorithms that exploit causal independence to
the induced width of the unmoralized Bayes network's graph.
It shows how causal independence can be utilized
within the general bucket-elimination framework for
belief updating, map and meu.
The paper builds upon previous work in that area.
-----------------------------------------------------------------------------
Rina Dechter dechter@ics.uci.edu
Information and Computer Science Dept. (714) 824-6556
University of California, Irvine fax: (714)-824-4056
Irvine, CA 92717