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.

Alternative uses of partitioning interwell tracer tests for DNAPL source zone characterisation

The heterogeneous nature of the subsurface and associated DNAPL morphologies often poses the greatest limitation to source zone clean-up strategies. Hence, detailed site characterisation techniques are required. The data presented in this paper has been collected from a series of laboratory 2-D tank experiments and numerical simulations of Partitioning Interwell Tracer Tests (PITT) in a wide range of aquifer conditions and DNAPL morphologies. Alternative uses of tracer breakthrough data have been developed In order to characterise the mass flux generated from the DNAPL source. By combining the laboratory and numerical data, a relationship between normalised mass flux and tracer-based average source zone DNAPL saturation has been established. Knowledge of such a relationship allows remediation targets to be identified, clean-up efficiencies to be evaluated, and increases the accuracy of any risk assessment.

Fracture and energy partitioning in uncooked and cooked noodles

Humans perform fascinating science experiments at home on a daily basis when they undertake the modification of natural and naturally-derived materials by a cooking process prior to consumption. The material properties of such foods are of interest to food scientists (texture is often fundamental to food acceptability), oral biologists (foods modulate feeding behavior), anthropologists (cooking is probably as old as the genus Homo and distinguishes us from all other creatures) and dentists (foods interact with tooth and tooth replacement materials). Materials scientists may be interested in the drastic changes in food properties observed over relatively short cooking times. In the current study, the mechanical properties of one of the most common (and oldest at 4,000+ years) foods on earth, the noodle, were examined as a function of cooking time. Two types of noodles were studied, each made from natural materials (wheat flour, salt, alkali and water) by kneading dough and passing them through a pasta-making machine. These were boiled for between 2-14 min and tested at regular intervals from raw to an overcooked state. Cyclic tensile tests at small strain levels were used to examine energy dissipation characteristics. Energy dissipation was >50% per cycle in uncooked noodles, but decreased by an order of magnitude with cooking. Fractional dissipation values remained approximately constant at cooking times greater than 7 min. Overall, a greater effect of cooking was on viscoplastic dissipation characteristics rather than on fracture resistance. The results of the current study plot the evolution of a viscoplastic mixture into an essentially elastic material in the space of 7 minutes and have broad implications for understanding what cooking does to food materials. In particular, they suggest that textural assessment by consumers of the optimally cooked state of food has a definite physical definition. © 2007 Materials Research Society.

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.