Semi-supervised model adaptation for statistical machine translation

Statistical machine translation systems are usually trained on large amounts of bilingual text (used to learn a translation model), and also large amounts of monolingual text in the target language (used to train a language model). In this article we explore the use of semi-supervised model adaptation methods for the effective use of monolingual data from the source language in order to improve translation quality. We propose several algorithms with this aim, and present the strengths and weaknesses of each one. We present detailed experimental evaluations on the French–English EuroParl data set and on data from the NIST Chinese–English large-data track. We show a significant improvement in translation quality on both tasks.     

Fast and accurate pricing of discretely monitored barrier options by numerical path integration

Barrier options are financial derivative contracts that are activated or deactivated according to the crossing of specified barriers by an underlying asset price. Exact models for pricing barrier options assume continuous monitoring of the underlying dynamics, usually a stock price. Barrier options in traded markets, however, nearly always assume less frequent observation, e.g. daily or weekly. These situations require approximate solutions to the pricing problem. We present a new approach to pricing such discretely monitored barrier options that may be applied in many realistic situations. In particular, we study daily monitored up-and-out call options of the European type with a single underlying stock. The approach is based on numerical approximation of the transition probability density associated with the stochastic differential equation describing the stock price dynamics, and provides accurate results in less than one second whenever a contract expires in a year or less. The flexibility of the method permits more complex underlying dynamics than the Black and Scholes paradigm, and its relative simplicity renders it quite easy to implement.     

Brokers and bricoleurs: entrepreneurship in Wales' online music scene

 The power of some new entrants to the music industry derives from their position as brokers in computer-mediated environments. Brokers act instrumentally to exploit their position within a network which, in turn, depends on their ability to build and sustain links (and, in computer-mediated environments, hyperlinks). Bricolage in computer-mediated entrepreneurship refers to the intuitive manipulation of resources in order to achieve (perhaps tacit) goals. Without careful stewardship of the new intellectual wealth thus created, bricolage may profit neither the individual nor the wider community. Accepted 14 March 2002     

Sintering of an ultra pure α-alumina powder: I. Densification,grain growth and sintering path

Sintering in air of an ultra pure α-alumina powder has been investigated. Isothermal experiments have been conducted on green samples shaped by slip casting. The grain growth and densification kinetics have been established. The “relative density/grain size” trajectory, called “sintering path”, has been drawn. Hypotheses concerning the mechanisms controlling grain growth and densification have been formulated. For the first time, it is shown that grain growth and densification kinetics exhibit two distinct regimes, where an initial point defect formation step plays a key role. When point defects have been generated, the diffusion of the associated Al³⁺ cations controls grain growth and densification.     

Quality of manual data collection in Java software: an empirical investigation

Data collection, both automatic and manual, lies at the heart of all empirical studies. The quality of data collected from software informs decisions on maintenance, testing and wider issues such as the need for system re-engineering. While of the two types stated, automatic data collection is preferable, there are numerous occasions when manual data collection is unavoidable. Yet, very little evidence exists to assess the error-proneness of the latter. Herein, we investigate the extent to which manual data collection for Java software compared with its automatic counterpart for the same data. We investigate three hypotheses relating to the difference between automated and manual data collection. Five Java systems were used to support our investigation. Results showed that, as expected, manual data collection was error-prone, but nowhere near the extent we had initially envisaged. Key indicators of mistakes in manual data collection were found to be poor developer coding style, poor adherence to sound OO coding principles, and the existence of relatively large classes in some systems. Some interesting results were found relating to the collection of public class features and the types of error made during manual data collection. The study thus offers an insight into some of the typical problems associated with collecting data manually; more significantly, it highlights the problems that poorly written systems have on the quality of visually extracted data.     

Two topographic maps for data visualisation

We review a new form of self-organizing map which is based on a nonlinear projection of latent points into data space, identical to that performed in the Generative Topographic Mapping (GTM) [Bishop et al. (1997) Neurl Comput 10(1): 215–234]. But whereas the GTM is an extension of a mixture of experts, our new model is an extension of a product of experts [Hinton (2000) Technical report GCNU TR 2000-004, Gatsby Computational Neuroscience Unit, University College, London]. We show visualisation results on some real data sets and compare with the GTM. We then introduce a second mapping based on harmonic averages and show that it too creates a topographic mapping of the data. We compare these mappings on real and artificial data sets. Responsible editor: Soumen Chakrabarti.     

Environment as a first class abstraction in multiagent systems

The current practice in multiagent systems typically associates the environment with resources that are external to agents and their communication infrastructure. Advanced uses of the environment include infrastructures for indirect coordination, such as digital pheromones, or support for governed interaction in electronic institutions. Yet, in general, the notion of environment is not well defined. Functionalities of the environment are often dealt with implicitly or in an ad hoc manner. This is not only poor engineering practice, it also hinders engineers to exploit the full potential of the environment in multiagent systems. In this paper, we put forward the environment as an explicit part of multiagent systems.We give a definition stating that the environment in a multiagent system is a first-class abstraction with dual roles: (1) the environment provides the surrounding conditions for agents to exist, which implies that the environment is an essential part of every multiagent system, and (2) the environment provides an exploitable design abstraction for building multiagent system applications. We discuss the responsibilities of such an environment in multiagent systems and we present a reference model for the environment that can serve as a basis for environment engineering. To illustrate the power of the environment as a design abstraction, we show how the environment is successfully exploited in a real world application. Considering the environment as a first-class abstraction in multiagent systems opens up new horizons for research and development in multiagent systems.     

A globally convergent algorithm for transportation continuous network design problem

The continuous network design problem (CNDP) is characterized by a bilevel programming model, in which the upper level problem is generally to minimize the total system cost under limited expenditure, while at the lower level the network users make choices with regard to route conditions following the user equilibrium principle. In this paper, the bilevel programming model for CNDP is transformed into a single level convex programming problem by virtue of an optimal-value function tool and the relationship between System Optimum (SO) and User Equilibrium (UE). By exploring the inherent nature of the CNDP, the optimal-value function for the lower level user equilibrium problem is proved to be continuously differentiable and its derivative in link capacity enhancement can be obtained efficiently by implementing user equilibrium assignment subroutine. However, the reaction (or response) function between the upper and lower level problem is implicit and its gradient is difficult to obtain. Although, here we approximately express the gradient with the difference concept at each iteration, based on the method of successive averages (MSA), we propose a globally convergent algorithm to solve the single level convex programming problem. Comparing with widely used heuristic algorithms, such as sensitivity analysis based (SAB) method, the proposed algorithm needs not strong hypothesis conditions and complex computation for the inverse matrix. Finally, a numerical example is presented to compare the proposed method with some existing algorithms.