We investigate the task of constraining NMT with pre-specified translations, which has practical significance for a number of research and industrial applications. Existing works impose pre-specified translations as lexical constraints during decoding, which are based on word alignments derived from target-to-source attention weights. However, multiple recent studies have found that word alignment derived from generic attention heads in the Transformer is unreliable. We address this problem by introducing a dedicated head in the multi-head Transformer architecture to capture external supervision signals. Results on five language pairs show that our method is highly effective in constraining NMT with pre-specified translations, consistently outperforming previous methods in translation quality.
Abstract. Remote Sensing (RS) techniques are increasingly used in urban tree inventory measurements for their improved accuracy and promptness over the conventional methods. The focus of this study is to evaluate the application of iPad Pro 2020 and its LiDAR sensor for urban trees reconstruction and Diameter at Breast Height (DBH) measurements. Altogether, 101 trees were scanned. We have used individual- and multiple-tree scan modes with different settings (Resolution: 10 mm, 15 mm, 20 mm; Confidence: High, Low). With these methods and settings, we have established 12 combinations. The 3DScannerAPP was used to scan and generate point clouds and to estimate DBH circle-fitting algorithm was used within the DendroCloud software. Among 12 methods, the only method with 10 mm resolution, high confidence, and multiple-tree mode has not achieved a 100% detection rate (97%). For multiple-tree mode, the highest estimation accuracy was 7.52% of relative RMSE, and for single-tree mode, it was 7.27%. Low confidence setting had significantly higher accuracy of DBH estimation than high confidence. Furthermore, single-tree mode had a significantly higher accuracy of DBH estimation than multiple-tree mode. The most efficient combination for DBH estimation of urban trees using 3DScannerAPP within iPad Pro 2020, when time and accuracy is considered, was multiple-tree mode with 15 mm resolution and low confidence.
Carrots are a type of vegetable with high nutrition. Before entering the market, the surface defect detection and sorting of carrots can greatly improve food safety and quality. To detect defects on the surfaces of carrots during combine harvest stage, this study proposed an improved knowledge distillation network structure that took yolo-v5s as the teacher network and a lightweight network that replaced the backbone network with mobilenetv2 and completed channel pruning as a student network (mobile-slimv5s). To make the improved student network adapt to the image blur caused by the vibration of the carrot combine harvester, we put the ordinary dataset Dataset (T) and dataset Dataset (S), which contains motion blurring treatment, into the teacher network and the improved lightweight network, respectively, for learning. By connecting multi-stage features of the teacher network, knowledge distillation was carried out, and different weight values were set for each feature to realize that the multi-stage features of the teacher network guide the single-layer output of the student network. Finally, the optimal lightweight network mobile-slimv5s was established, with a network model size of 5.37 MB. The experimental results show that when the learning rate is set to 0.0001, the batch size is set to 64, and the dropout is set to 0.65, the model accuracy of mobile-slimv5s is 90.7%, which is significantly higher than other algorithms. It can synchronously realize carrot harvesting and surface defect detection. This study laid a theoretical foundation for applying knowledge distillation structures to the simultaneous operations of crop combine harvesting and surface defect detection in a field environment. This study effectively improves the accuracy of crop sorting in the field and contributes to the development of smart agriculture.
While neural models based on the Transformer architecture achieve the State-of-the-Art translation performance, it is well known that the learned target-to-source attentions do not correlate well with word alignment. There is an increasing interest in inducing accurate word alignment in Transformer, due to its important role in practical applications such as dictionary-guided translation and interactive translation. In this article, we extend and improve the recent work on unsupervised learning of word alignment in Transformer on two dimensions: a) parameter initialization from a pre-trained cross-lingual language model to leverage large amounts of monolingual data for learning robust contextualized word representations, and b) regularization of the training objective to directly model characteristics of word alignments which results in favorable word alignments receiving more concentrated probabilities. Experiments on benchmark data sets of three language pairs show that the proposed methods can significantly reduce alignment error rate (AER) by at least 3.7 to 7.7 points on each language pair over two recent works on improving the Transformer's word alignment. Moreover, our methods can achieve better alignment results than GIZA++ on certain test sets.
Epidemiological studies suggest that higher serum uric acid (sUA) level is significantly associated with nonalcoholic fatty liver disease (NAFLD) development. However, little information is available on the relationships between sUA and NAFLD remission. In the present study, 841 NAFLD males (30-75 years) were recruited from a Chinese prospective cohort study (PMMJS) and followed up for five years. The baseline sUA levels of participants were categorized into four quartiles: 191 μmol/L≤ sUA ≤ 347 μmol/L, 347 μmol/L < sUA ≤ 392 μmol/L, 392 μmol/L < sUA ≤ 441 μmol/L and 441 μmol/L
Atherosclerosis is, in part, caused by immune and inflammatory cell infiltration into the vascular wall, leading to enhanced inflammation and lipid accumulation in the aortic endothelium. Understanding the molecular mechanisms underlying this disease is critical for the development of new therapies. Our recent studies demonstrate that epsins, a family of ubiquitin-binding endocytic adaptors, are critical regulators of atherogenicity. Given the fundamental contribution lesion macrophages make to fuel atherosclerosis, whether and how myeloid-specific epsins promote atherogenesis is an open and significant question.
Abstract Background Uterine cervical cancer (UCC) was the fourth leading cause of cancer death among women worldwide. The conventional MRI hardly revealing the microstructure information. This study aimed to compare the value of amide proton transfer-weighted imaging (APTWI) and diffusion kurtosis imaging (DKI) in evaluating the histological grade of cervical squamous carcinoma (CSC) in addition to routine diffusion-weighted imaging (DWI). Methods Forty-six patients with CSC underwent pelvic DKI and APTWI. The magnetization transfer ratio asymmetry (MTRasym), apparent diffusion coefficient (ADC), mean diffusivity (MD) and mean kurtosis (MK) were calculated and compared based on the histological grade. Correlation coefficients between each parameter and histological grade were calculated. Results The MTRasym and MK values of grade 1 (G1) were significantly lower than those of grade 2 (G2), and those parameters of G2 were significantly lower than those of grade 3 (G3). The MD and ADC values of G1 were significantly higher than those of G2, and those of G2 were significantly higher than those of G3. MTRasym and MK were both positively correlated with histological grade (r = 0.789 and 0.743, P < 0.001), while MD and ADC were both negatively correlated with histological grade (r = − 0.732 and - 0.644, P < 0.001). For the diagnosis of G1 and G2 CSCs, AUC (APTWI+DKI + DWI) > AUC (DKI + DWI) > AUC (APTWI+DKI) > AUC (APTWI+DWI) > AUC (MTRasym) > AUC (MK) > AUC (MD) > AUC (ADC), where the differences between AUC (APTWI+DKI + DWI), AUC (DKI + DWI) and AUC (ADC) were significant. For the diagnosis of G2 and G3 CSCs, AUC (APTWI+DKI + DWI) > AUC (APTWI+DWI) > AUC (APTWI+DKI) > AUC (DKI + DWI) > AUC (MTRasym) > AUC (MK) > AUC (MD > AUC (ADC), where the differences between AUC (APTWI+DKI + DWI), AUC (APTWI+DWI) and AUC (ADC) were significant. Conclusion Compared with DWI and DKI, APTWI is more effective in identifying the histological grades of CSC. APTWI is recommended as a supplementary scan to routine DWI in CSCs.
'/%3e%3cuse xlink:href='%23a' transform='rotate(23.036 .093 25.536)'/%3e%3cuse xlink:href='%23a' transform='rotate(45.87 1.273 16.18)'/%3e%3cuse xlink:href='%23a' transform='rotate(69.945 .996 12.078)'/%3e%3cuse xlink:href='%23a' transform='rotate(20.66 -19.689 31.932)'/%3e%3c/svg%3e)
