Compression techniques applied to multiple speech recognition systems

Speech recognition systems typically contain many Gaussian distributions, and hence a large number of parameters. This makes them both slow to decode speech, and large to store. Techniques have been proposed to decrease the number of parameters. One approach is to share parameters between multiple Gaussians, thus reducing the total number of parameters and allowing for shared likelihood calculation. Gaussian tying and subspace clustering are two related techniques which take this approach to system compression. These techniques can decrease the number of parameters with no noticeable drop in performance for single systems. However, multiple acoustic models are often used in real speech recognition systems. This paper considers the application of Gaussian tying and subspace compression to multiple systems. Results show that two speech recognition systems can be modelled using the same number of Gaussians as just one system, with little effect on individual system performance. Copyright © 2009 ISCA.

A probabilistic energy model for non-domestic building sectors applied to analysis of school buildings in greater London

The diversity of non-domestic buildings at urban scale poses a number of difficulties to develop models for large scale analysis of the stock. This research proposes a probabilistic, engineering-based, bottom-up model to address these issues. In a recent study we classified London's non-domestic buildings based on the service they provide, such as offices, retail premise, and schools, and proposed the creation of one probabilistic representational model per building type. This paper investigates techniques for the development of such models. The representational model is a statistical surrogate of a dynamic energy simulation (ES) model. We first identify the main parameters affecting energy consumption in a particular building sector/type by using sampling-based global sensitivity analysis methods, and then generate statistical surrogate models of the dynamic ES model within the dominant model parameters. Given a sample of actual energy consumption for that sector, we use the surrogate model to infer the distribution of model parameters by inverse analysis. The inferred distributions of input parameters are able to quantify the relative benefits of alternative energy saving measures on an entire building sector with requisite quantification of uncertainties. Secondary school buildings are used for illustrating the application of this probabilistic method. © 2012 Elsevier B.V. All rights reserved.

Local stress relaxation and shear banding in a dry foam under shear.

We have developed a realistic simulation of 2D dry foams under quasistatic shear. After a short transient, a shear-banding instability is observed. These results are compared with measurements obtained on real 2D (confined) foams. The numerical model allows us to exhibit the mechanical response of the material to a single plastication event. From the analysis of this elastic propagator, we propose a scenario for the onset and stability of the flow localization process in foams, which should remain valid for most athermal amorphous systems under creep flow.

Separating poroviscoelastic deformation mechanisms in hydrogels

Hydrogels have applications in drug delivery, mechanical actuation, and regenerative medicine. When hydrogels are deformed, load-relaxation arising from fluid flow - poroelasticity - and from rearrangement of the polymer network - viscoelasticity - is observed. The physical mechanisms are different in that poroelastic relaxation varies with experimental length-scale while viscoelastic does not. Here, we show that poroviscoelastic load-relaxation is the product of the two individual responses. The difference in length-scale dependence of the two mechanisms can be exploited to uniquely determine poroviscoelastic properties from simultaneous analysis of multi-scale indentation experiments, providing insight into hydrogel physical behavior. © 2013 American Institute of Physics.

Travel behaviour applied in freight transportation using intelligent products

The route planning problem for an order in freight transportation involves the selection of the best route for its transportation given a set of options that the network can offer. In its adaptive (or dynamic) version, the problem deals with the planning of a new route for an order while it is actually in transit typically because part or all of its pre-selected route is blocked or disrupted. In the intelligent product approach we are proposing, an order would be capable of identifying and evaluating such new routes in an automated manner and choosing the most preferable one without the intervention of humans. Because such approaches seek to mirror (and then automate) human decision making, in this paper we seek to identify new ways for dynamic route planning in industrial logistics inspired by the way people make similar decisions about their journey when they travel in multi-modal networks. We propose a new simulation game as a methodological tool for capturing their travel behaviour and we use it in this study. The results show that a simulation game can be used for capturing strategies and tactics of travellers and that intelligent products can provide a proper platform for the usage of such strategies in freight logistics. © 2012 IEEE.

Convex relaxation of mixture regression with efficient algorithms

We develop a convex relaxation of maximum a posteriori estimation of a mixture of regression models. Although our relaxation involves a semidefinite matrix variable, we reformulate the problem to eliminate the need for general semidefinite programming. In particular, we provide two reformulations that admit fast algorithms. The first is a max-min spectral reformulation exploiting quasi-Newton descent. The second is a min-min reformulation consisting of fast alternating steps of closed-form updates. We evaluate the methods against Expectation-Maximization in a real problem of motion segmentation from video data.

Shielding efficiency and E(J) characteristics measured on large melt cast Bi-2212 hollow cylinders in axial magnetic fields

We show that tubes of melt cast Bi-2212 used as current leads for LTS magnets can also act as efficient magnetic shields. The magnetic screening properties under an axial DC magnetic field are characterized at several temperatures below the liquid nitrogen temperature (77 K). Two main shielding properties are studied and compared with those of Bi-2223, a material that has been considered in the past for bulk magnetic shields. The first property is related to the maximum magnetic flux density that can be screened, Blim; it is defined as the applied magnetic flux density below which the field attenuation measured at the centre of the shield exceeds 1000. For a cylinder of Bi-2212 with a wall thickness of 5 mm and a large ratio of length over radius, Blim is evaluated to 1 T at T = 10 K. This value largely exceeds the Blim value measured at the same temperature on similar tubes of Bi-2223. The second shielding property that is characterized is the dependence of Blim with respect to variations of the sweep rate of the applied field, dBapp/dt. This dependence is interpreted in terms of the power law E = Ec(J/Jc)^n and allows us to determine the exponent n of this E(J) characteristics for Bi-2212. The characterization of the magnetic field relaxation involves very small values of the electric field. This gives us the opportunity to experimentally determine the E(J) law in an unexplored region of small electric fields. Combining these results with transport and AC shielding measurements, we construct a piecewise E(J) law that spans over 8 orders of magnitude of the electric field.