A Novel Beam Search to Improve Neural Machine Translation for English-Chinese

Neural Machine Translation (NMT) is an end-to-end learning approach for automated translation, overcoming the weaknesses of conventional phrase-based translation systems. Although NMT based systems have gained their popularity in commercial translation applications, there is still plenty of room for improvement. Being the most popular search algorithm in NMT, beam search is vital to the translation result. However, traditional beam search can produce duplicate or missing translation due to its target sequence selection strategy. Aiming to alleviate this problem, this paper proposed neural machine translation improvements based on a novel beam search evaluation function. And we use reinforcement learning to train a translation evaluation system to select better candidate words for generating translations. In the experiments, we conducted extensive experiments to evaluate our methods. CASIA corpus and the 1,000,000 pairs of bilingual corpora of NiuTrans are used in our experiments. The experiment results prove that the proposed methods can effectively improve the English to Chinese translation quality.     

视频侦查中人像智能分析应用及算法优化

人像智能分析指的是对视频或录像中的人像进行结构化和可视化分析,对目标人物进行性别、年龄、发型等特征的智能识别,这项技术在视频侦查中有极高的应用价值。人像识别早期的算法是通过人工提取特征,通过学习低级视觉特征来针对不同属性进行分类学习,这种基于传统方法的模型表现常常不尽如人意。在计算机视觉领域,通过海量图像数据学习的神经网络比传统方法有更丰富的信息量和特征可以被提取。文章尝试通过深度学习技术训练神经网络模型对行人进行检测和识别,对于衣着不同的行人进行智能识别,具有更好的鲁棒性,提升了视频人像识别的准确率,拓展了人工智能技术在身份识别领域的应用。     

冷轧预变形对新型铝锂合金的时效析出与力学性能影响

采用大气熔炼铸造及热变形方法制备了Al-4.5Cu-1Li-0.7Mg-1Zn-0.3Ag-0.3Mn-0.2Zr新型铝锂合金板材。通过维氏硬度、拉伸性能、扫描电镜、透射电镜等方法，研究了固溶后不同冷轧预变形量对显微组织和力学性能影响。结果表明，时效前的冷轧预变形量有效促进了新型铝锂基体合金中T1(Al2CuLi)相的析出与均匀分布，减少了θ′(Al2Cu)相的体积分数。冷轧预变形量的增加，缩短了峰时效时间，晶界析出相由连续析出变为非连续析出，无沉淀析出区宽度变小。当冷轧预变形量为15%时，时效态合金的屈服强度与抗拉强度分别达到了668 MPa、690 MPa，延伸率保持在7.9%。     

D-limonene Inhibits Pentylenetetrazole-Induced Seizure via Adenosine A2A Receptor Modulation on GABAergic Neuronal Activity

Background: Epilepsy is a chronic neurological disorder characterized by the recurrence of seizures. One-third of patients with epilepsy may not respond to antiseizure drugs. Purpose: We aimed to examine whether D-limonene, a cyclic monoterpene, exhibited any antiseizure activity in the pentylenetetrazole (PTZ)-induced kindling mouse model and in vitro. Methods: PTZ kindling mouse model was established by administering PTZ (30 mg/kg) intraperitoneally to mice once every 48 h. We performed immunoblot blots, immunohistochemistry (IHC), and high-performance liquid chromatography (HPLC) analysis after the behavioral study. Results: An acute injection of PTZ (60 mg/kg) induced seizure in mice, while pretreatment with D-limonene inhibited PTZ-induced seizure. Repeated administration of PTZ (30 mg/kg) increased the seizure score gradually in mice, which was reduced in D-limonene (10 mg/kg)-pretreated group. In addition, D-limonene treatment increased glutamate decarboxylase-67 (GAD-67) expression in the hippocampus. Axonal sprouting of hippocampal neurons after kindling was inhibited by D-limonene pretreatment. Moreover, D-limonene reduced the expression levels of Neuronal PAS Domain Protein 4 (Npas4)-induced by PTZ. Furthermore, the adenosine A2A antagonist {"type":"entrez-protein","attrs":{"text":"SCH58261","term_id":"1052882304","term_text":"SCH58261"}}SCH58261 and ZM241385 inhibited anticonvulsant activity and gamma-aminobutyric acid (GABA)ergic neurotransmission-induced by D-limonene. Conclusion: These results suggest that D-limonene exhibits anticonvulsant activity through modulation of adenosine A2A receptors on GABAergic neuronal function.     

The effect of optimized carbon source on the synthesis and composition of exopolysaccharides produced by Lactobacillus paracasei

This study aimed to predict the optimal carbon source for higher production of exopolysaccharides (EPS) by Lactobacillus paracasei TD 062, and to evaluate the effect of this carbon source on the production and monosaccharide composition of EPS. We evaluated the EPS production capacity of 20 strains of L. paracasei under the same conditions. We further investigated L. paracasei TD 062, which showed the highest EPS-producing activity (0.609 g/L), by examining the associated biosynthesis pathways for EPS. Genomics revealed that fructose, mannose, trehalose, glucose, galactose, and lactose were carbon sources that L. paracasei TD 062 could use to produce EPS. We identified an EPS synthesis gene cluster that could participate in transport, export, and sugar chain synthesis, and generate 6 sugar nucleotides. Experimental results showed that the sugar content of the EPS produced using fermentation with the optimized carbon source (fructose, mannose, trehalose, glucose, galactose, and lactose) increased by 115%. Furthermore, use of the optimized carbon source changed the monosaccharide content of the associated EPS. The results of enzyme activity measurements showed significant increases in the activity of 2 key enzymes involved in the glycoside synthesis pathway. Our study revealed that optimizing the carbon source provided for fermentation not only increased the production of EPS, but also affected the composition of the monosaccharides by increasing enzyme activity in the underlying synthesis pathways, suggesting an important role for carbon source in the production of EPS by L. paracasei TD 062.     

基于TRIZ理论滤咀排列鼓工装夹具优化设计

针对滤咀排列鼓工装夹具所存在的制造难度大,精度难于保证,效率低的问题。利用TRIZ理论进行分析,根据技术矛盾解决方法并结合问题的实际情况,获得了具体可行的解决方案。经过实践证明,此设计方案是可行的、可靠的,大大缩短了夹具的加工周期,提高了生产效率,节约了成本。     

基于深度卷积神经网络的视觉里程计研究

视觉里程计利用视频信息来估计相机运动的位姿参数，实现对智能体的定位。传统视觉里程计方法需要特征提取、特征匹配/跟踪、外点剔除、运动估计、优化等流程，解算非常复杂，因此，提出了基于卷积神经网络的方法来实现端对端的单目视觉里程计。借助卷积神经网络对彩色图片自动学习提取图像帧间变化的全局特征，将用于分类的卷积神经网络转化为帧间时序特征网络，通过三层全连接层输出相机的帧间相对位姿参数。在KITTI数据集上的实验结果表明，提出的Deep-CNN-VO模型可以较准确地估计车辆的运动轨迹，证明了方法的可行性。在简化了复杂模型的基础上，与传统的视觉里程计系统相比，该模型的精度也有所提高。     

Single-Crystalline TiO2(B) Nanobelts with Unusual Large Exposed {100} Facets and Enhanced Li-Storage Capacity

The {100} facet of single-crystalline TiO₂(B) is an ideal platform for inserting Li ions, but it is hard to be obtained due to its high surface energy. Here, the single-crystalline TiO₂(B) nanobelts from H₂Ti₃O₇ with nearly 70% {100} facets exposed are synthesized, which significantly enhances Li-storage capacity. The first-principle calculations demonstrate an ab in-plane 2D diffusion through the exposed {100} facets. As a consequence, the nanobelts can significantly accommodate Li ions in LiTiO₂ formula with specific capacity up to 335 mAh g⁻¹, which is in good agreement with the electrochemical characterizations. Coating with conductive and protective poly(3,4-ethylenedioxythiophene)-poly(styrenesulfonate), the cut-off discharge voltage is as low as 0.5 V, leading to a capacity of 160.7 mAh g⁻¹ after 1500 cycles with a retention rate of 66% at 1C. This work provides a practical strategy to increase the Li-ion capacity and cycle stability by tailoring the crystal orientation and nanostructures.     

掘进巷道快速过断层施工技术应用

基于四候煤矿3108回风顺槽过F3断层期间,顶板出现严重破碎、下沉现象,提出了俯斜台阶法施工工艺,并对断层破碎带顶板采取MF-2型化学材料注浆加固以及"钢筋锚索网+U29梯形棚"联合支护措施。支护效果检验结果表明,联合支护有效控制了断层破碎区顶板下沉、破碎现象,保证了巷道顶板稳定性。