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. 相似文献
Most of the commonly used hydrological models do not account for the actual evapotranspiration (ETa) as a key contributor to water loss in semi-arid/arid regions. In this study, the HEC-HMS (Hydrologic Engineering Center Hydrologic Modeling System) model was calibrated, modified, and its performance in simulating runoff resulting from short-duration rainfall events was evaluated. The model modifications included integrating spatially distributed ETa, calculated using the surface energy balance system (SEBS), into the model. Evaluating the model’s performance in simulating runoff showed that the default HEC-HMS model underestimated the runoff with root mean squared error (RMSE) of 0.14 m3/s (R2?=?0.92) while incorporating SEBS ETa into the model reduced RMSE to 0.01 m3/s (R2?=?0.99). The integration of HECHMS and SEBS resulted in smaller and more realistic latent heat flux estimates translated into a lower water loss rate and a higher magnitude of runoff simulated by the HECHMS model. The difference between runoff simulations using the default and modified model translated into an average of 95,000 m3 runoff per rainfall event (equal to seasonal water requirement of ten-hectare winter wheat) that could be planned and triggered for agricultural purposes, flood harvesting, and groundwater recharge in the region. The effect of ETa on the simulated runoff volume is expected to be more pronounced during high evaporative demand periods, longer rainfall events, and larger catchments. The outcome of this study signifies the importance of implementing accurate estimates of evapotranspiration into a hydrological model.
Piles are widely applied to substructures of various infrastructural buildings. Soil has a complex nature; thus, a variety of empirical models have been proposed for the prediction of the bearing capacity of piles. The aim of this study is to propose a novel artificial intelligent approach to predict vertical load capacity of driven piles in cohesionless soils using support vector regression (SVR) optimized by genetic algorithm (GA). To the best of our knowledge, no research has been developed the GA-SVR model to predict vertical load capacity of driven piles in different timescales as of yet, and the novelty of this study is to develop a new hybrid intelligent approach in this field. To investigate the efficacy of GA-SVR model, two other models, i.e., SVR and linear regression models, are also used for a comparative study. According to the obtained results, GA-SVR model clearly outperformed the SVR and linear regression models by achieving less root mean square error (RMSE) and higher coefficient of determination (R2). In other words, GA-SVR with RMSE of 0.017 and R2 of 0.980 has higher performance than SVR with RMSE of 0.035 and R2 of 0.912, and linear regression model with RMSE of 0.079 and R2 of 0.625.
In the present work, layer thickness of duplex coating made from thermo-reactive deposition and diffusion has been predicted by Adaptive network-based fuzzy inference systems (ANFIS). A duplex surface treatment on five steels has been developed involving nitrocarburizing and followed by chromium thermo-reactive deposition (TRD) techniques. The TRD process was performed in molten salt bath at 550, 625 and 700 °C for 1–30 h. The process formed a thickness up to 9.5 μm of chromium carbonitride coatings on a hardened diffusion zone. A model based on ANFIS for predicting the layer thickness of duplex coating of the specimens has been presented. To build the model, training and testing using experimental results from 84 specimens were conducted. The data used as inputs in ANFIS models are arranged in a format of twelve parameters that cover the chemical composition (C, Mn, Si, Cr, Mo, V, W), the pre-nitriding time, ferro-chromium particle size, ferro-chromium weight percent, salt bath temperature and coating time. According to these input parameters, in the Adaptive network-based fuzzy inference system models, the layer thickness of duplex coating of each specimen was predicted. The training and testing results in ANFIS models have shown a strong potential for predicting the layer thickness of duplex coating. 相似文献
Construction activities within the wetted‐perimeter of rivers, referred to as in‐stream construction, are prevalent economically and environmentally motivated activities having direct interactions with sensitive lotic environments. Currently, there is a paucity of research related to in‐stream construction activities and their effects on aquatic ecosystems. In‐stream construction‐induced suspended sediment may result in harmful effects to aquatic flora and fauna. Regulatory frameworks worldwide focus primarily on concentration, with limited consideration for duration and no consideration for spatial extents of suspended sediment exposures. This research develops theoretical concentration, duration, and spatial extent exposure risk relationships for riverine ecosystems to demonstrate the influence of each mechanism during typical in‐stream construction activities. To reduce exposure risk, concentration and duration may be considered pragmatically, based on anticipated activity characteristics and site conditions. Spatial exposure characteristics are important to consider, as illustrated by our finding that activities located near the channel centerline may result in greater exposure risk than similar activities conducted near the streambank. Current regulatory frameworks worldwide do not sufficiently consider all exposure risk mechanisms present during in‐stream construction‐induced suspended sediment releases, possibly inhibiting efforts to reduce adverse environmental effects. This research improves our understanding of suspended sediment in lotic environments and may help environmental managers better evaluate and manage in‐stream construction activities. 相似文献
In this work, the transport properties of gaseous penetrant through several dense glassy polymeric membranes are studied. The nonequilibrium lattice fluid (NELF) in conjunction with the modified Fick's law and dual mode sorption model was used to simulate the gas transport in glassy polymeric membranes. The approach is based on the sorption, diffusion, in which solubility is calculated based on the NELF model, and diffusion coefficient is obtained from the product thermodynamic coefficient and molecular mobility. The governing equation is solved by the finite element method using COMSOL multi-physics software. The developed model for gas permeability of glassy polymeric membrane can be applied in a wide range of pressure and temperature. The comparison of the calculated permeability and solubility of gasses with the experimental data represented the ability of the developed model. Increasing feed gas temperature increases the gas permeability, while this variation leads to lower gas solubility in the glassy polymeric membranes. The effect of feed temperature and pressure on permeability and solubility is investigated, and the experimental data from literature are described by the developed model. A good prediction of the experimental data can be observed over the considered condition. 相似文献
Recently, by defining suitable fuzzy temporal logics, temporal properties of dynamic systems are specified during model checking process, yet a few numbers of fuzzy temporal logics along with capable corresponding models are developed and used in system design phase, moreover in case of having a suitable model, it suffers from the lack of a capable model checking approach. Having to deal with uncertainty in model checking paradigm, this paper introduces a fuzzy Kripke model (FzKripke) and then provides a verification approach using a novel logic called Fuzzy Computation Tree Logic* (FzCTL*). Not only state space explosion is handled using well-known concepts like abstraction and bisimulation, but an approximation method is also devised as a novel technique to deal with this problem. Fuzzy program graph, a generalization of program graph and FzKripke, is also introduced in this paper in consideration of higher level abstraction in model construction. Eventually modeling, and verification of a multi-valued flip-flop is studied in order to demonstrate capabilities of the proposed models. 相似文献
Blasting operation is widely used method for rock excavation in mining and civil works. Ground vibration and air-overpressure (AOp) are two of the most detrimental effects induced by blasting. So, evaluation and prediction of ground vibration and AOp are essential. This paper presents a new combination of artificial neural network (ANN) and K-nearest neighbors (KNN) models to predict blast-induced ground vibration and AOp. Here, this combination is abbreviated using ANN-KNN. To indicate performance of the ANN-KNN model in predicting ground vibration and AOp, a pre-developed ANN as well as two empirical equations, presented by United States Bureau of Mines (USBM), were developed. To construct the mentioned models, maximum charge per delay (MC) and distance between blast face and monitoring station (D) were set as input parameters, whereas AOp and peak particle velocity (PPV), as a vibration index, were considered as output parameters. A database consisting of 75 datasets, obtained from the Shur river dam, Iran, was utilized to develop the mentioned models. In terms of using three performance indices, namely coefficient correlation (R2), root mean square error and variance account for, the superiority of the ANN-KNN model was proved in comparison with the ANN and USBM equations. 相似文献
This paper proposes a new vocal-based emotion recognition method using random forests, where pairs of the features on the whole speech signal, namely, pitch, intensity, the first four formants, the first four formants bandwidths, mean autocorrelation, mean noise-to-harmonics ratio and standard deviation, are used in order to recognize the emotional state of a speaker. The proposed technique adopts random forests to represent the speech signals, along with the decision-trees approach, in order to classify them into different categories. The emotions are broadly categorised into the six groups, which are happiness, fear, sadness, neutral, surprise, and disgust. The Surrey Audio-Visual Expressed Emotion database is used. According to the experimental results using leave-one-out cross-validation, by means of combining the most significant prosodic features, the proposed method has an average recognition rate of \(66.28\%\), and at the highest level, the recognition rate of \(78\%\) has been obtained, which belongs to the happiness voice signals. The proposed method has \(13.78\%\) higher average recognition rate and \(28.1\%\) higher best recognition rate compared to the linear discriminant analysis as well as \(6.58\%\) higher average recognition rate than the deep neural networks results, both of which have been implemented on the same database. 相似文献