Multi-task learning has been shown to be effective in various applications, including discriminative SMT. We present an experimental evaluation of the question whether multi-task learning depends on a “natural” division of data into tasks that balance shared and individual knowledge, or whether its inherent regularization makes multi-task learning a broadly applicable remedy against overfitting...

Minimum Error Rate Training (MERT) remains one of the preferred methods for tuning linear parameters in machine translation systems, yet it faces significant issues. First, MERT is an unregularized learner and is therefore prone to overfitting. Second, it is commonly used on a noisy, non-convex loss function that becomes more difficult to optimize as the number of parameters increases. To addre...

We study the impact of source length and verbosity of the tuning dataset on the performance of parameter optimizers such as MERT and PRO for statistical machine translation. In particular, we test whether the verbosity of the resulting translations can be modified by varying the length or the verbosity of the tuning sentences. We find that MERT learns the tuning set verbosity very well, while P...

As the speed gap between CPU and I/O is getting wider and wider, I/O latency plays a more important role to the overall system performance than it used to be. Prefetching consecutive data blocks within individual files has been explored to hide, at least partially, the I/O latency. Unfortunately, this does not reduce the inter-file latency occurring when a program reads new files during its exe...

We describe the line search used in the minimum error rate training algorithm (Och, 2003) as the “inside score” of a weighted proof forest under a semiring defined in terms of wellunderstood operations from computational geometry. This conception leads to a straightforward complexity analysis of the dynamic programming MERT algorithms of Macherey et al. (2008) and Kumar et al. (2009) and practi...

Central banks and other forecasters are increasingly interested in various aspects of density forecasts. However, recent sharp changes in macroeconomic volatility – such as the Great Moderation and the more recent sharp rise in volatility associated with greater variation in energy prices and the deep global recession – pose significant challenges to density forecasting. Accordingly, this paper...

We propose a comparative performance evaluation of security protocols. The novelty of our approach lies in the use of a polynomial mathematical model that captures the performance of classes of cryptographic algorithms instead of capturing the performance of each algorithm separately, approach that is used in other papers. A major advantage of using such a model is that it does not require impl...

Methods Persistent asthmatics ≥ 6 years old under GINA outpatient treatment were consecutively included in a real life design and a society perspective/bottom-up approach. They underwent routine clinical visits at 3to 4-month intervals and 2 interviews with 6-month intervals using a structured instrument, made by researchers not involved in their clinical treatment. Data on asthma costs were co...

Integrated simulation systems are important tools to decrease the damage caused by various disasters. In peacetime, it would be used for disaster-proof city planning, disaster prevention plans, and disaster response training. At the time of a disaster occurs, it will be used for real-time simulation to predict disasters. The paper describes the simulation system that has the following features:...

Structured perceptrons are attractive due to their simplicity and speed, and have been used successfully for tuning the weights of binary features in a machine translation system. In attempting to apply them to tuning the weights of real-valued features with highly skewed distributions, we found that they did not work well. This paper describes a modification to the update step and compares the...