dear colleagues, due to technical problems the following paper
T.Kocka and R.Castelo 'Improved Learning of Bayesian Networks' published in this last year's UAI conference proceedings had all plots printed wrongly. we thank the organisers for providing quickly an errata page at the conference, but we would like to warn those of you that did not attend the conference that the plots in the proceedings are not right (there are lines missing). for that reason we let you know that you can download the pdf, .ps or .ps zipped version of the paper (where the plots are fine) through the following urls: http://www.cs.uu.nl/~roberto/uai2k1.pdf http://www.cs.uu.nl/~roberto/uai2k1.ps http://www.cs.uu.nl/~roberto/uai2k1.zip the paper introduces a new and simple learning policy for Bayesian Networks, and shows, among other things, that can recover the Alarm network perfectly in about 50 local transformations of the empty DAG (taking about 2 minutes in a current PC) using a hill-climber without any causal ordering. thanks, robert castelo & tomas kocka.
