Use of contexts in language model interpolation and adaptation

Language models (LMs) are often constructed by building multiple individual component models that are combined using context independent interpolation weights. By tuning these weights, using either perplexity or discriminative approaches, it is possible to adapt LMs to a particular task. This paper investigates the use of context dependent weighting in both interpolation and test-time adaptation of language models. Depending on the previous word contexts, a discrete history weighting function is used to adjust the contribution from each component model. As this dramatically increases the number of parameters to estimate, robust weight estimation schemes are required. Several approaches are described in this paper. The first approach is based on MAP estimation where interpolation weights of lower order contexts are used as smoothing priors. The second approach uses training data to ensure robust estimation of LM interpolation weights. This can also serve as a smoothing prior for MAP adaptation. A normalized perplexity metric is proposed to handle the bias of the standard perplexity criterion to corpus size. A range of schemes to combine weight information obtained from training data and test data hypotheses are also proposed to improve robustness during context dependent LM adaptation. In addition, a minimum Bayes' risk (MBR) based discriminative training scheme is also proposed. An efficient weighted finite state transducer (WFST) decoding algorithm for context dependent interpolation is also presented. The proposed technique was evaluated using a state-of-the-art Mandarin Chinese broadcast speech transcription task. Character error rate (CER) reductions up to 7.3 relative were obtained as well as consistent perplexity improvements. © 2012 Elsevier Ltd. All rights reserved.

Energy use of buildings at urban scale: A case study of london school buildings

The diversity of non-domestic buildings at urban scale poses a number of difficulties to develop building stock models. This research proposes an engineering-based bottom-up stock model in a probabilistic manner to address these issues. School buildings are used for illustrating the application of this probabilistic method. Two sampling-based global sensitivity methods are used to identify key factors affecting building energy performance. The sensitivity analysis methods can also create statistical regression models for inverse analysis, which are used to estimate input information for building stock energy models. The effects of different energy saving measures are analysed by changing these building stock input distributions.

The incentives for supply chain collaboration to improve material efficiency in the use of steel: An analysis using input output techniques

In the face of increasing demand and limited emission reduction opportunities, the steel industry will have to look beyond its process emissions to bear its share of emission reduction targets. One option is to improve material efficiency - reducing the amount of metal required to meet services. In this context, the purpose of this paper is to explore why opportunities to improve material efficiency through upstream measures such as yield improvement and lightweighting might remain underexploited by industry. Established input-output techniques are applied to the GTAP 7 multi-regional input-output model to quantify the incentives for companies in key steel-using sectors (such as property developers and automotive companies) to seek opportunities to improve material efficiency in their upstream supply chains under different short-run carbon price scenarios. Because of the underlying assumptions, the incentives are interpreted as overestimates. The principal result of the paper is that these generous estimates of the incentives for material efficiency caused by a carbon price are offset by the disincentives to material efficiency caused by labour taxes. Reliance on a carbon price alone to deliver material efficiency would therefore be misguided and additional policy interventions to support material efficiency should be considered. © 2013 Elsevier B.V.

The use of an MgB2 hollow cylinder and pulse magnetized (RE)BCO bulk for magnetic levitation applications

Both MgB2 and (RE)BCO bulk materials can provide a highly compact source of magnetic field when magnetized. The properties of these materials when magnetized by a pulsed field are potentially useful for a number of applications, including magnetic levitation. This paper reports on pulsed field magnetization of single 25 mm diameter (RE)BCO bulks using a recently constructed pulse magnetization facility, which allows an automated sequence of pulses to be delivered. The facility allows measurement of force between a magnetized (RE)BCO bulk and a bulk MgB2 hollow cylinder, which is field cooled in the field of the magnetized (RE)BCO bulk. Hysteresis cycling behavior for small displacement is also measured to extract the stiffness value. The levitation forces up to 500 N were obtained, the highest ever measured between two bulks and proves the concept of a bulk-bulk superconducting bearing design. © 2002-2011 IEEE.

Use of 2G coated conductors for efficient shielding of DC magnetic fields

This paper reports the results of an experimental investigation of the performance of two types of magnetic screens assembled from YBa2Cu3O7-d (YBCO) coated conductors. Since effective screening of the axial DC magnetic field requires the unimpeded flow of an azimuthal persistent current, we demonstrate a configuration of a screening shell made out of standard YBCO coated conductor capable to accomplish that. The screen allows the persistent current to flow in the predominantly azimuthal direction at a temperature of 77 K. The persistent screen, incorporating a single layer of superconducting film, can attenuate an external magnetic field of up to 5 mT by more than an order of magnitude. For comparison purposes, another type of screen which incorporates low critical temperature quasi-persistent joints was also built. The shielding technique we describe here appears to be especially promising for the realization of large scale high-Tc superconducting screens.

Use of a high-temperature superconducting coil for magnetic energy storage

A high temperature superconducting magnetic energy storage device (SMES) has been realised using a 350 m-long BSCCO tape wound as a pancake coil. The coil is mounted on a cryocooler allowing temperatures down to 17.2 K to be achieved. The temperature dependence of coil electrical resistance R(T) shows a superconducting transition at T 102.5 K. Measurements of the V(I) characteristics were performed at several temperatures between 17.2 K and 101.5 K to obtain the temperature dependence of the critical current (using a 1 νV/cm criterion). Critical currents were found to exceed 100 A for T < 30 K. An electronic DC-DC converter was built in order to control the energy flow in and out of the superconducting coil. The converter consists of a MOS transistor bridge switching at a 80 kHz frequency and controlled with standard Pulse Width Modulation (PWM) techniques. The system was tested using a 30 V squared wave power supply as bridge input voltage. The coil current, the bridge input and output voltages were recorded simultaneously. Using a 10 A setpoint current in the superconducting coil, the whole system (coil + DC-DC converter) can provide a stable output voltage showing uninterruptible power supply (UPS) capabilities over 1 s. © 2006 IOP Publishing Ltd.

The use of liquid crystals as electrorheological fluids in microsystems: Model and measurements

Fluids with controllable flow properties have gained considerable interest in the past few years. Some of these fluids such as magnetorheologic fluids are now widely applied to active dampers and valves. Although these fluids show promising properties for microsystems, their applicability is limited to the microscale since particles suspended in these fluids tend to obstruct microchannels. This paper investigates the applicability of electrorheologic liquid crystals (LCs) in microsystems. Since LCs do not contain suspended particles, they show intrinsic advantages over classic rheologic fluids in micro-applications. This paper presents a novel physical model that describes the static and the dynamic behaviour of electrorheologic LCs. The developed model is validated by comparing simulations and measurements performed on a rectangular microchannel. This assessment shows that the model presented in this paper is able to simulate both static and dynamic properties accurately. Therefore, this model is useful for the understanding, simulation and optimization of devices using LCs as electrorheological fluid. In addition, measurements performed in this paper reveal remarkable properties of LCs, such as high bandwidths and high changes in flow resistance. © 2006 IOP Publishing Ltd.

An example of the use of neural computing techniques in materials science - The modelling of fatigue thresholds in Ni-base superalloys

Two adaptive numerical modelling techniques have been applied to prediction of fatigue thresholds in Ni-base superalloys. A Bayesian neural network and a neurofuzzy network have been compared, both of which have the ability to automatically adjust the network's complexity to the current dataset. In both cases, despite inevitable data restrictions, threshold values have been modelled with some degree of success. However, it is argued in this paper that the neurofuzzy modelling approach offers real benefits over the use of a classical neural network as the mathematical complexity of the relationships can be restricted to allow for the paucity of data, and the linguistic fuzzy rules produced allow assessment of the model without extensive interrogation and examination using a hypothetical dataset. The additive neurofuzzy network structure means that redundant inputs can be excluded from the model and simple sub-networks produced which represent global output trends. Both of these aspects are important for final verification and validation of the information extracted from the numerical data. In some situations neurofuzzy networks may require less data to produce a stable solution, and may be easier to verify in the light of existing physical understanding because of the production of transparent linguistic rules. © 1999 Elsevier Science S.A.

On the use of predictor-corrector method for coupled Monte Carlo burnup codes

This paper investigates the effect of the burnup coupling scheme on the numerical stability and accuracy of coupled Monte-Carlo depletion calculations. We show that in some cases, even the Predictor Corrector method with relatively short time steps can be numerically unstable. In addition, we present two possible extensions to the Euler predictor-corrector (PC) method, which is typically used in coupled burnup calculations. These modifications allow using longer time steps, while maintaining numerical stability and accuracy. © 2013 Elsevier Ltd. All rights reserved.

The potential use of 241Am as proliferation resistant burnable poison in PWRs

This paper discusses the use of 241Am as proliferation resistant burnable poison for light water reactors. Homogeneous addition of small (as little as 0.12%) amounts of 241Am to the conventional light water reactor fuel results in significant increase in 238Pu/Pu ratio in the discharged fuel improving its proliferation resistance. Moreover, 241Am, admixed to the fuel, acts as burnable absorber allowing for substantial reduction in conventional reactivity control means without a notable fuel cycle length penalty. This is possible due to favorable characteristics of 241Am transmutation chain. The fuel cycle length penalty of introducing 241Am into the core is evaluated and discussed, as well as the impact of He production in the fuel pins and degradation of reactivity feedback coefficients. Proliferation resistance and reactivity control features related to the use of 241Am are compared to those of using 237Np, which has also been suggested as an additive to the conventional fuel in order to improve its proliferation resistance. It was found that 241Am admixture is more favorable than 237Np admixture because of the smaller fuel cycle length penalty and higher burnable poison savings. Addition of either 237Np or 241Am would provide substantial but not ultimate protection from misuse of Pu originating in the spent fuel from the commercial power reactors. Therefore, the benefits from application of the concept would have to be carefully evaluated against the additional costs and proliferation risks associated with manufacturing of 237Np or 241Am doped fuel. Although this work concerns specifically with PWRs, the conclusions could also be applied to BWRs and, to some extent, to other thermal spectrum reactor types. © 2009 Elsevier Ltd. All rights reserved.

The potential use of 241Am as proliferation resistant burnable poison

This paper discusses the use of 141Am as proliferation resistant burnable poison for light water reactors. Homogeneous addition of small (less than 1 %) amounts of 241Am to the conventional LWR fuel results in significant increase in 238Pu/Pu ratio in the discharged fuel improving its proliferation resistance. Moreover, 241Am, admixed to the fuel, acts as burnable absorber allowing for substantial reduction in conventional reactivity control means without notable fuel cycle length penalty. This is possible due to favourable characteristics of 241Am transmutation chain. The fuel cycle length penalty of introducing 241Am into the core is evaluated and discussed, as well as the impact of He production in the fuel pins and degradation of reactivity feedback coefficients. Proliferation resistance and reactivity control features related to the use of 241Am are compared to those of using 237Np, which has also been suggested as an additive to the conventional fuel in order to improve its proliferation resistance. It was found that 241Am admixture is more favourable than 237Np admixture because of the smaller fuel cycle length penalty and higher burnable poison savings.

Use of axially graded burnable boron for hot-spot temperature reduction in a pressurized water reactor core

Shortly after the loading of a pressurized water reactor (PWR) core, the axial power distribution in fresh fuel has already attained the characteristic, almost flat shape, typical of burned fuel. At beginning of cycle (BOC), however, the axial distribution is centrally peaked. In assemblies hosting uniform burnable boron rods, this BOC peaking is even more pronounced. A reduction in the axial peaking is today often achieved by shortening the burnable boron rods by some 30 cm at each edge. It is shown that a two-zone grading of the boron rod leads, in a representative PWR cycle, to a reduction of the hot-spot temperature of approximately 70 °C, compared with the nongraded case. However, with a proper three-zone grading of the boron rod, an additional 20 °C may be cut off the hot-spot temperature. Further, with a slightly skewed application of this three-zone grading, an additional 50 °C may be cut off. The representative PWR cycle studied was cycle 11 of the Indian Point 2 station, with a simplification in the number of fuel types and in the burnup distribution. The analysis was based on a complete three-dimensional burnup calculation. The code system was ELCOS, with BOXER as an assembly code for the generation of burnup-dependent cross sections and SILWER as a three-dimensional core code with thermal-hydraulic feedback.