Pitman-Yor Diffusion Trees

We introduce the Pitman Yor Diffusion Tree (PYDT) for hierarchical clustering, a generalization of the Dirichlet Diffusion Tree (Neal, 2001) which removes the restriction to binary branching structure. The generative process is described and shown to result in an exchangeable distribution over data points. We prove some theoretical properties of the model and then present two inference methods: a collapsed MCMC sampler which allows us to model uncertainty over tree structures, and a computationally efficient greedy Bayesian EM search algorithm. Both algorithms use message passing on the tree structure. The utility of the model and algorithms is demonstrated on synthetic and real world data, both continuous and binary.

Gaussian Process Regression Networks

We introduce a new regression framework, Gaussian process regression networks (GPRN), which combines the structural properties of Bayesian neural networks with the non-parametric flexibility of Gaussian processes. This model accommodates input dependent signal and noise correlations between multiple response variables, input dependent length-scales and amplitudes, and heavy-tailed predictive distributions. We derive both efficient Markov chain Monte Carlo and variational Bayes inference procedures for this model. We apply GPRN as a multiple output regression and multivariate volatility model, demonstrating substantially improved performance over eight popular multiple output (multi-task) Gaussian process models and three multivariate volatility models on benchmark datasets, including a 1000 dimensional gene expression dataset.

Sleeping sites, sleeping trees, and sleep-related behaviors of black crested gibbons (Nomascus concolor jingdongensis) at Mt. Wuliang, central Yunnan, China

Data on sleep-related behaviors were collected for a group of central Yunnan black crested gibbons (Nomascus concolor jingdongensis) at Mt. Wuliang, Yunnan, China from March 2005 to April 2006. Members of the group usually formed four sleeping units (adult male and juvenile, adult female with one semi-dependent black infant, adult female with one dependent yellow infant, and subadult male) spread over different sleeping trees. Individuals or units preferred specific areas to sleep; all sleeping sites were situated in primary forest, mostly (77%) between 2,200 and 2,400 m in elevation. They tended to sleep in the tallest and thickest trees with large crowns on steep slopes and near important food patches. Factors influencing sleeping site selection were (1) tree characteristics, (2) accessibility, and (3) easy escape. Few sleeping trees were used repeatedly by the same or other members of the group. The gibbons entered the sleeping trees on average 128 min before sunset and left the sleeping trees on average 33 min after sunrise. The lag between the first and last individual entering the trees was on average 17.8 min. We suggest that sleep-related behaviors are primarily adaptations to minimize the risk of being detected by predators. Sleeping trees may be chosen to make approach and attack difficult for the predator, and to provide an easy escape route in the dark. In response to cold temperatures in a higher habitat, gibbons usually sit and huddle together during the night, and in the cold season they tend to sleep on ferns and/or orchids.

Statistical parametric speech synthesis based on speaker and language factorization

An increasingly common scenario in building speech synthesis and recognition systems is training on inhomogeneous data. This paper proposes a new framework for estimating hidden Markov models on data containing both multiple speakers and multiple languages. The proposed framework, speaker and language factorization, attempts to factorize speaker-/language-specific characteristics in the data and then model them using separate transforms. Language-specific factors in the data are represented by transforms based on cluster mean interpolation with cluster-dependent decision trees. Acoustic variations caused by speaker characteristics are handled by transforms based on constrained maximum-likelihood linear regression. Experimental results on statistical parametric speech synthesis show that the proposed framework enables data from multiple speakers in different languages to be used to: train a synthesis system; synthesize speech in a language using speaker characteristics estimated in a different language; and adapt to a new language. © 2012 IEEE.