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.

The nature and origin of deep ocean clay crust from the Gulf of Guinea

Deep ocean sediments off the west coast of Africa exhibit a peculiar undrained strength profile in the form of a crust, albeit of exceptionally high water content, overlying normally consolidated clay. Hot-oil pipelines are installed into these crustal sediments, so their origins and characteristics are of great interest to pipeline designers. This paper provides evidence for the presence of burrowing invertebrates in crust material, and for the way sediment properties are modified through their creation of burrows, and through the deposition of faecal pellets. A variety of imaging techniques are used to make these connections, including photography, scanning electron microscopy and X-ray computer tomography. However, the essential investigative technology is simply the wet-sieving of natural cores, which reveals that up to 60% by dry mass of the crustal material can consist of smooth, highly regular, sand-sized capsules that have been identified as the faecal pellets of invertebrates such as polychaetes. Mechanical tests reveal that these pellets are quite robust under effective stresses of the order of 10 kPa, acting like sand grains within a matrix of fines. Their abundance correlates closely with the measured strength of the crust. While this can easily be accepted in the context of a pellet fraction as high as 60%, the question arises how a smaller proportion of pellets, such as 20%, is apparently able to enhance significantly the strength of a sediment that otherwise appears to be normally consolidated. A hypothesis is suggested based on the composition of the matrix of fines around the pellets. These appear to consist of agglomerates of clay platelets, which may be the result of the breakdown of pellets by other organisms. Their continued degradation at depths in excess of 1 m is taken to explain the progressive loss of crustal strength thereafter.