Dear CLASS-L people, Hi. I e-mail you now for a few reasons. One is to announce a tribute to Chris Wallace (1933-2004), whose (Computer J., 1968) paper saw both the beginning of the Minimum Message Length (MML) principle and of MML mixture modelling. I'd also like to announce an apparent uniqueness result for log-loss scoring of probabilistic classification, and I'd finally like to mention the existence of a postdoc. fellowship.
Re Chris Wallace (1933-2004), the Computer Journal (Oxford Univ. Press) has now placed online on its WWW site all the articles from the (``soon'') forthcoming special issue in honour and memory of C. S. Wallace (1933 - 2004). The articles are interspersed in various places throughout the list at http://comjnl.oxfordjournals.org/papbyrecent.dtl , and they include Murray A. Jorgensen and Geoffrey J. McLachlan Wallace's Approach to Unsupervised Learning: The Snob Program www.doi.org : 10.1093/comjnl/bxm121 (on Chris Wallace's work on MML mixture modelling) and David L. Dowe Foreword re C. S. Wallace www.doi.org : 10.1093/comjnl/bxm117 (in which I attempt to survey Chris's career and his work, with a particular focus on Minimum Message Length [MML] and achievements). Re probabilistic classification, footnote 175 of the above article (http://comjnl.oxfordjournals.org/cgi/content/full/bxm117#FN175) gives an apparent uniqueness result suggesting advantages and merits of log-loss over the likes of Area Under Curve (AUC) and ROC, etc. (This article also discusses issues of relative performances of (Bayesian) MML and (e.g.) Akaike's AIC as a form of penalized likelihood.) And, re postdocs, David Dowe is looking for a postdoctoral fellow for collaborative work on (Bayesian) Minimum Message Length (MML) (and/or Kolmogorov complexity) applied to any of (e.g.) clustering, mixture modelling, hierarchical clustering and hierarchical mixture modelling, latent factor analysis, regression, time series, DNA microarray analysis, decision trees, decision graphs, support vector machines (SVMs), classification, Bayesian nets, phylogenetic networks (or phylogenetic graphs), Inductive Logic Programming (ILP) and hybrids thereof, etc. The relevant researcher should have between two and eight years of post-doctoral experience as of September 2008. A successful applicant will be appointed as Research Fellow (Level B) beginning January 2009, with a salary per annum up to Aus$82,951 (including superannuation) and Aus$50,000 additional research support for 5 years. More specific and detailed information is at www.csse.monash.edu.au/~dld/2009_DDowe_MML_MonashFellowships-guidelines.pdf . The stated University deadline is Tuesday 5 August 2008, but Monash's Faculty of I.T. will have a much earlier internal deadline of approx. 2ish weeks earlier - so we should try to aim for that. I would welcome any help you can provide in helping me to find a prospective applicant. Yours sincerely, David Dowe (www.csse.monash.edu.au/~dld). ---------------------------------------------- CLASS-L list. Instructions: http://www.classification-society.org/csna/lists.html#class-l
