Dear colleagues, We are pleased to announce the release of PRISM2.2 which is available for download at:
http://rjida.meijo-u.ac.jp/prism/ PRISM is a probabilistic extension of Prolog designed for a unified interface between users and machine learning tasks. The user declaratively writes probabilistic model as PRISM programs and applies to them a rich array of built-in predicates for exact probability computation, Viterbi inference, five types of parameter learning (EM/MAP,DAEM,VT,VB,VB-VT) and Metropolis-Hastings sampling for Bayesian inference. The new version further extends functionalities of the previous version with two innovative features: * capability of computing an infinite sum of probabilities which makes it possible to deal with probabilistic models with cyclic structure including Markov chains and affixes in PCFGs, and * capability of computing and learning conditional random fields (CRFs) by which a large class of CRFs including logistic-regression, linear-chain CRFs and CRF-CFGs become usable. PRISM2.2 is now a unique platform covering generative models such as BNs, HMMs and PCFGs and powerful discriminative models such as linear-chain CRFs. With Best Regards, Taisuke Sato, Yoshitaka Kameya, Ryosuke Kojima, Neng-Fa Zhou _______________________________________________ uai mailing list uai@ENGR.ORST.EDU https://secure.engr.oregonstate.edu/mailman/listinfo/uai
