Tree-Structured Stick Breaking Processes for Hierarchical Data

Many data are naturally modeled by an unobserved hierarchical structure. In this paper we propose a flexible nonparametric prior over unknown data hierarchies. The approach uses nested stick-breaking processes to allow for trees of unbounded width and depth, where data can live at any node and are infinitely exchangeable. One can view our model as providing infinite mixtures where the components have a dependency structure corresponding to an evolutionary diffusion down a tree. By using a stick-breaking approach, we can apply Markov chain Monte Carlo methods based on slice sampling to perform Bayesian inference and simulate from the posterior distribution on trees. We apply our method to hierarchical clustering of images and topic modeling of text data.

A model for the erosive wear of rubber at oblique impact angles

A model has been developed to predict the erosive wear behaviour of elastomers under conditions of glancing impact by small hard particles. Previous work has shown the erosive wear mechanism of elastomers under these conditions to be similar in nature to that of abrasive wear by a sharp blade. The model presented here was developed from the model of Southern and Thomas for sliding abrasion, by combining their treatment of the growth of surface cracks with a model for particle impact in which the force - displacement relationship for an idealized flat-ended punch on a semi-infinite elastic solid was assumed. In this way an expression for the erosive wear rate was developed, and compared with experimental measurements of wear rate for natural rubber, styrene - butadiene rubber and a highly crosslinked polybutadiene rubber. Good qualitative agreement was found between the predictions of the model and the experimental measurements. The variation of erosion rate with impact velocity, impact angle, particle size, elastic modulus of the material, coefficient of friction and fatigue properties were all well accounted for. Quantitative agreement was less good, and the effects of erosive particle shape could not be accounted for. The reasons for these discrepancies are discussed. © 1992 IOP Publishing Ltd.

Comparative molecular cytogenetic studies in the order Carnivora: mapping chromosomal rearrangements onto the phylogenetic tree

We have made a set of chromosome-specific painting probes for the American mink by degenerate oligonucleotide primed-PCR (DOP-PCR) amplification of flow-sorted chromosomes. The painting probes were used to delimit homologous chromosomal segments among human, red fox, dog, cat and eight species of the family Mustelidae, including the European mink, steppe and forest polecats, least weasel, mountain weasel, Japanese sable, striped polecat, and badger. Based on the results of chromosome painting and G-banding, comparative maps between these species have been established. The integrated map demonstrates a high level of karyotype conservation among mustelid species. Comparative analysis of the conserved chromosomal segments among mustelids and outgroup species revealed 18 putative ancestral autosomal segments that probably represent the ancestral chromosomes, or chromosome arms, in the karyotype of the most recent ancestor of the family Mustelidae. The proposed 2n = 38 ancestral Mustelidae karyotype appears to have been retained in some modern mustelids, e.g., Martes, Lutra, ktonyx, and Vormela. The derivation of the mustelid karyotypes from the putative ancestral state resulted from centric fusions, fissions, the addition of heterochromatic arms, and occasional pericentric inversions. Our results confirm many of the evolutionary conclusions suggested by other data and strengthen the topology of the carnivore phylogenetic tree through the inclusion of genome-wide chromosome rearrangements. Copyright (C) 2002 S. KargerAG, Basel.

Characterization of three fibrinogenolytic enzymes from Chinese green tree viper (Trimeresurus stejnegeri) venom

Rong Gao, Yun Zhang, Qing-Xiong Meng, Wen-Hui Lee, Dong-Sheng Li, Yu-liang Xiong and Wan-Yu Wang. Characterization of three fibrinogenolytic enzymes from Chinese green tree viper (Trimeresurus stejneger ) venom. Toxicon 36, 457-467, 1998.-From the venom of Chinese green tree viper (Trimeresurus stejnegeri), three distinct fibrinogenolytic enzymes: stejnefibrase-l, stejnefibrase-2 and stejnefibrase-3, were purified by gel filtration, ion-exchange chromatography and reverse-phase high-performance chromatograghy (HPLC). SDS-PAGE analysis of those three enzymes showed that they consisted of a single polypeptide chain with mel. wt of -50 000, 31 000 and 32 000, respectively. Like TSV-PA (a specific plasminogen activator) and stejnobin (a fibrinogen-clotting enzyme) purified from the same venom, stejnfibrase-1, -2 and -3 were able to hydrolyze several chromogenic substrate. On the other hand, different from TSV-PA. and stejnobin, stejnefibrase-l, -2 and -3 did not activate plasminogen and did not possess fibrinogen-clotting activity. The three purified enzymes directly degraded fibrinogen to small fragments and rendered it unclottable by thrombin. Stejnefibrase-2 degraded preferentially BE-chain while stejnefibrase-l and -3 cleaved concomitantly Ax and B beta-chains of fibrinogen. None of these proteases degraded the gamma-chain of fibrinogen. When correlated with the loss of clottability of fibrinogen, the most active enzyme was stejnefibrase-l. The activities of the three enzymes were inhibited by phenylmethylsulfonyl fluoride (PMSF) and p-nitrophenyl-p-guanidinobenzoate (NPGB), indicating that like TSV-PA and stejnobin, they are venom serine proteases. (C) 1998 Elsevier Science Ltd. All rights reserved.

Decision tree-based context clustering based on cross validation and hierarchical priors

The standard, ad-hoc stopping criteria used in decision tree-based context clustering are known to be sub-optimal and require parameters to be tuned. This paper proposes a new approach for decision tree-based context clustering based on cross validation and hierarchical priors. Combination of cross validation and hierarchical priors within decision tree-based context clustering offers better model selection and more robust parameter estimation than conventional approaches, with no tuning parameters. Experimental results on HMM-based speech synthesis show that the proposed approach achieved significant improvements in naturalness of synthesized speech over the conventional approaches. © 2011 IEEE.