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.

Mapping forces in 3D elastic assembly of grains

Our understanding of the elasticity and rheology of disordered materials, such as granular piles, foams, emulsions or dense suspensions relies on improving experimental tools to characterize their behaviour at the particle scale. While 2D observations are now routinely carried out in laboratories, 3D measurements remain a challenge. In this paper, we use a simple model system, a packing of soft elastic spheres, to illustrate the capability of X-ray microtomography to characterise the internal structure and local behaviour of granular systems. Image analysis techniques can resolve grain positions, shapes and contact areas; this is used to investigate the material's microstructure and its evolution upon strain. In addition to morphological measurements, we develop a technique to quantify contact forces and estimate the internal stress tensor. As will be illustrated in this paper, this opens the door to a broad array of static and dynamical measurements in 3D disordered systems

Mapping the conceptual design activity of interdisciplinary teams

The conceptual design phase of any project is, by its very nature, a vibrant, creative and dynamic period. It can also be disorganized with much backtracking accompanying the exchange of information between design team members. The transfer of information, ideas and opinion is critical to the development of concepts and as such, rather than being recognized as merely a component of conceptual design activity, it needs to be understood and, ultimately, managed. This paper describes an experimental workshop involving fifteen design professionals in which conceptual design activity was tracked, and subsequently mapped, in order to test and validate a tentative design framework (phase and activity model). The nature of the design progression of the various teams is captured and analyzed, allowing a number of conclusions to be drawn regarding both the iterative nature of this phase of design and how teams of professionals actually design together.

Developing a framework for mapping industrial emergence

The industrial landscape is becoming increasingly complex and dynamic, with innovative technologies stimulating the emergence of new industries and business models. This paper presents a preliminary framework for mapping industrial emergence, based on roadmapping principles, in order to understand the nature and characteristics of such phenomena. The focus at this stage is on historical examples of industrial emergence, with the preliminary framework based on observations from 20 'quick scan' maps, one of which is used to illustrate the framework. The learning from these historical cases, combined with further industrial consultation and literature review, will be used to develop practical methods for strategy and policy application. The paper concludes by summarising key learning points and further work needed to achieve these outcomes. © 2009 PICMET.

Constrained discriminative mapping transforms for unsupervised speaker adaptation

Discriminative mapping transforms (DMTs) is an approach to robustly adding discriminative training to unsupervised linear adaptation transforms. In unsupervised adaptation DMTs are more robust to unreliable transcriptions than directly estimating adaptation transforms in a discriminative fashion. They were previously proposed for use with MLLR transforms with the associated need to explicitly transform the model parameters. In this work the DMT is extended to CMLLR transforms. As these operate in the feature space, it is only necessary to apply a different linear transform at the front-end rather than modifying the model parameters. This is useful for rapidly changing speakers/environments. The performance of DMTs with CMLLR was evaluated on the WSJ 20k task. Experimental results show that DMTs based on constrained linear transforms yield 3% to 6% relative gain over MLE transforms in unsupervised speaker adaptation. © 2011 IEEE.

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.

Semi-supervised video segmentation using tree structured graphical models

We present a novel, implementation friendly and occlusion aware semi-supervised video segmentation algorithm using tree structured graphical models, which delivers pixel labels alongwith their uncertainty estimates. Our motivation to employ supervision is to tackle a task-specific segmentation problem where the semantic objects are pre-defined by the user. The video model we propose for this problem is based on a tree structured approximation of a patch based undirected mixture model, which includes a novel time-series and a soft label Random Forest classifier participating in a feedback mechanism. We demonstrate the efficacy of our model in cutting out foreground objects and multi-class segmentation problems in lengthy and complex road scene sequences. Our results have wide applicability, including harvesting labelled video data for training discriminative models, shape/pose/articulation learning and large scale statistical analysis to develop priors for video segmentation. © 2011 IEEE.

Genome-wide high-resolution mapping and functional analysis of DNA methylation in arabidopsis.

Cytosine methylation is important for transposon silencing and epigenetic regulation of endogenous genes, although the extent to which this DNA modification functions to regulate the genome is still unknown. Here we report the first comprehensive DNA methylation map of an entire genome, at 35 base pair resolution, using the flowering plant Arabidopsis thaliana as a model. We find that pericentromeric heterochromatin, repetitive sequences, and regions producing small interfering RNAs are heavily methylated. Unexpectedly, over one-third of expressed genes contain methylation within transcribed regions, whereas only approximately 5% of genes show methylation within promoter regions. Interestingly, genes methylated in transcribed regions are highly expressed and constitutively active, whereas promoter-methylated genes show a greater degree of tissue-specific expression. Whole-genome tiling-array transcriptional profiling of DNA methyltransferase null mutants identified hundreds of genes and intergenic noncoding RNAs with altered expression levels, many of which may be epigenetically controlled by DNA methylation.