We explore unsupervised language model adaptation techniques for Statistical Machine Translation. The hypotheses from the machine translation output are converted into queries at different levels of representation power and used to extract similar sentences from very large monolingual text collection. Specific language models are then build from the retrieved data and interpolated with a general background model. Experiments show significant improvements when translating with these adapted language models.

Lens-free digital in-line holography (LDIH) is a promising microscopic tool that can overcome the limitations (e.g., field of view) of traditional lens-based microcopy. Images produced by LDIH, however, require extensive computation time to reconstruct objet images from complex diffraction patterns, which limits LDIH utility for point-of-care applications, particularly in resource limited settings. Here, we describe a new deep-learning (DL) based approach to process LDIH images in the context of cellular analyses. Specifically, we captured hologram images of cells surface-labeled with molecular-specific microbeads, and trained neural networks to classify these holograms without reconstruction. Using raw holograms as input, the trained networks could group individual cells according to the number of beads attached. Particularly, VGG-19 pretrained networks showed robust performance even with noisy experimental data. Combined with the developed DL approach, LDIH could be realized as low-cost, portable tool for point-ofcare diagnostics.

Lens-free digital in-line holography (LDIH) is a promising microscopic tool that overcomes several drawbacks (e.g., limited field of view) of traditional lens-based microcopy. However, extensive computation is required to reconstruct object images from the complex diffraction patterns produced by LDIH. This limits LDIH utility for point-of-care applications, particularly in resource limited settings. We describe a deep transfer learning (DTL) based approach to process LDIH images in the context of cellular analyses. Specifically, we captured holograms of cells labeled with molecular-specific microbeads and trained neural networks to classify these holograms without reconstruction. Using raw holograms as input, the trained networks were able to classify individual cells according to the number of cell-bound microbeads. The DTL-based approach including a VGG19 pretrained network showed robust performance with experimental data. Combined with the developed DTL approach, LDIH could be realized as a low-cost, portable tool for point-of-care diagnostics.

It is inevitable that defects happen to key components of the long-running high-speed trains. Thus as an effective inspection approach for defects, image detection becomes significantly important for operation and maintenance in the railway industry. However, a massive number of images collected by inspection devices challenge traditional methods based on manual effort. To address this issue, this paper proposed an automatic detection method, termed as multi-stage pipeline for defect detection (MPDD). MPDD includes two stages, component detection stage improves RPN anchor mechanism and way of feature fusion to promote detection performance, defect classification stage combines super-resolution strategy with CNN to improve defect classification performance. Experiments on high-speed train defect dataset shown that MPDD can reach the highest mAP of 0.792. The mAP on NEU surface defect database reached to 0.765 at the speed of 203ms per image.

Syntactic word reordering is essential for translations across different grammar structures between syntactically distant languagepairs. In this paper, we propose to embed local and non-local word reordering decisions in a synchronous context free grammar, and leverages the grammar in a chartbased decoder. Local word-reordering is effectively encoded in Hiero-like rules; whereas non-local word-reordering, which allows for long-range movements of syntactic chunks, is represented in tree-based reordering rules, which contain variables correspond to sourceside syntactic constituents. We demonstrate how these rules are learned from parallel corpora. Our proposed shallow Tree-to-String rules show significant improvements in translation quality across different test sets.

No abstract

Abstract. A unit hydrograph (UH) obtained from past storms can be used to predict a direct runoff hydrograph (DRH) based on the effective rainfall hyetograph (ERH) of a new storm. The objective functions in commonly used linear programming (LP) formulations for obtaining an optimal UH are (1) minimizing the sum of absolute deviations (MSAD) and (2) minimizing the largest absolute deviation (MLAD). This paper proposes two alternative LP formulations for obtaining an optimal UH, namely, (1) minimizing the weighted sum of absolute deviations (MWSAD) and (2) minimizing the range of deviations (MRNG), In this paper the predicted DRHs as well as the regenerated DRHs by using the UHs obtained from different LP formulations were compared using a statistical crossvalidation technique. The golden section search method was used to determine the optimal weights for the model of MWSAD. The numerical results show that the UH by MRNG is better than that by MLAD in regenerating and predicting DRHs. It is also found that the model MWSAD with a properly selected weighing function would produce a UH that is better in predicting the DRHs than the commonly used MSAD. weight assigned to error associated with estimating en.error associated with over-estimation of Q n .error associated with under-estimation of Qn.

Unit hydrographs (UHs), along with design rainfalls, are frequently used to determine the discharge hydrograph for design and evaluation of hydraulic structures. Due to the presence of various uncertainties in its derivation, the resulting UH is inevitably subject to uncertainty. Consequently, the performance of hydraulic structures under the design storm condition is uncertain. This paper integrates the linearly constrained Monte-Carlo simulation with the UH theory and routing techniques to evaluate the reliability of hydraulic structures. The linear constraint is considered because the water volume of each generated design direct runoff hydrograph should be equal to that of the design effective rainfall hyetograph or the water volume of each generated UH must be equal to one inch (or cm) over the watershed, For illustration, the proposed methodology is applied to evaluate the overtopping risk of a hypothetical flood detention reservoir downstream of Tong-Tou watershed in Taiwan.