Learning the Structure of Deep Sparse Graphical Models

Deep belief networks are a powerful way to model complex probability distributions. However, learning the structure of a belief network, particularly one with hidden units, is difficult. The Indian buffet process has been used as a nonparametric Bayesian prior on the directed structure of a belief network with a single infinitely wide hidden layer. In this paper, we introduce the cascading Indian buffet process (CIBP), which provides a nonparametric prior on the structure of a layered, directed belief network that is unbounded in both depth and width, yet allows tractable inference. We use the CIBP prior with the nonlinear Gaussian belief network so each unit can additionally vary its behavior between discrete and continuous representations. We provide Markov chain Monte Carlo algorithms for inference in these belief networks and explore the structures learned on several image data sets.

Dem analysis of soil-pipeline interaction in sand under lateral and upward movements at deep embedment

The soil-pipeline interactions under lateral and upward pipe movements in sand are investigated using DEM analysis. The simulations are performed for both medium and dense sand conditions at different embedment ratios of up to 60. The comparison of peak dimensionless forces from the DEM and earlier FEM analyses shows that, for medium sand, both methods show similar peak dimensionless forces. For dense sand, the DEM analysis gives more gradual transition of shallow to deep failure mechanisms than the FEM analysis and the peak dimensionless forces at very deep depth are higher in the DEM analysis than in the FEM analysis. Comparison of the deformation mechanism suggests that this is due to the differences in soil movements around the pipe associated with its particulate nature. The DEM analysis provides supplementary data of the soil-pipeline interaction in sand at deep embedment condition.

Analysis of case histories on deep excavations in marine clay

Two case histories on deep excavation of marine clay are used to study the use of a decision-making tool based on a new deign method called the Mobilized Strength Design (MSD) method which allows the designer to use a simple method of predicting ground displacements during deep excavation. This application can approximately satisfy both safety and serviceability requirements by predicting stresses and displacements under working conditions by introducing the concept of "Mobilizable soil strength". The new method accommodates a number of features which are important to design of underground construction between retaining walls, including different deformation mechanism in different stages of excavation. The influence of wall depth, wall flexibility and stratified ground are the major focus of this paper. These developments should make it possible for a design engineer to take informed decisions on the influence of wall stiffness, or on the need for a jet-grouted base slab, for example, without having to conduct project-specific Finite Element Analysis.