Chinese word segmentation with local and global context representation learning

A local and global context representation learning model for Chinese characters is designed and a Chinese word segmentation method based on character representations is proposed in this paper. First, the proposed Chinese character learning model uses the semanties of loeal context and global context to learn the representation of Chinese characters. Then, Chinese word segmentation model is built by a neural network, while the segmentation model is trained with the eharaeter representations as its input features. Finally, experimental results show that Chinese charaeter representations can effectively learn the semantic information. Characters with similar semantics cluster together in the visualize space. Moreover, the proposed Chinese word segmentation model also achieves a pretty good improvement on precision, recall and f-measure.

Rethinking Global Context in Crowd Counting

This paper investigates the role of global context for crowd counting.Specifically,a pure transformer is used to extract features with global information from overlapping image patches.Inspired by classification,we add a context token to the input sequence,to facilitate information exchange with tokens corresponding to image patches throughout transformer layers.Due to the fact that transformers do not explicitly model the tried-and-true channel-wise interactions,we propose a token-attention module(TAM)to recalibrate encoded features through channel-wise attention informed by the context token.Beyond that,it is adopted to predict the total person count of the image through regression-token module(RTM).Extensive experiments on various datasets,including ShanghaiTech,UCFQNRF,JHU-CROWD++and NWPU,demonstrate that the proposed context extraction techniques can significantly improve the performanceover the baselines.

Intercultural Trust in Global Contexts:Synthesizing a Western Nomological Approach with a Chinese Systems Approach

Intercultural trust in global contexts plays a central role in helping people from different cultures to communicate comfortably,which is essential for cooperation.Attempting to construct a framework that might foster international cooperation,and thus be helpful for coping with global emergencies,we relate a Western nomological approach to an Eastern systems approach to analyse intercultural trust in global contexts.Considering cultural impacts on intercultural trust and the nomological framework of cultural differences,we propose an intercultural trust model to interpret how cultural differences influence trust.A qualitative study of Chinese-Irish interactions was conducted to interpret this model.We organized 10 seminars on intercultural trust,and interviewed 16 people to further explore the respondents'deeper feelings and experiences about intercultural trust in global contexts.Through this study,we have identified factors impacting on intercultural trust,and found that intercultural trust can be developed and improved in various ways.To llustrate these ways,we have provided tactics and methods for building intercultural trust in global contexts.Implications are highlighted for organizations to avoid cultural clashes and relevant political or economic risks.

Natural Image Matting with Attended Global Context

Image matting is to estimate the opacity of foreground objects from an image. A few deep learning based methods have been proposed for image matting and perform well in capturing spatially close information. However, these methods fail to capture global contextual information, which has been proved essential in improving matting performance. This is because a matting image may be up to several megapixels, which is too big for a learning-based network to capture global contextual information due to the limit size of a receptive field. Although uniformly downsampling the matting image can alleviate this problem, it may result in the degradation of matting performance. To solve this problem, we introduce a natural image matting with the attended global context method to extract global contextual information from the whole image, and to condense them into a suitable size for learning-based network. Specifically, we first leverage a deformable sampling layer to obtain condensed foreground and background attended images respectively. Then, we utilize a contextual attention layer to extract information related to unknown regions from condensed foreground and background images generated by a deformable sampling layer. Besides, our network predicts a background as well as the alpha matte to obtain more purified foreground, which contributes to better qualitative performance in composition. Comprehensive experiments show that our method achieves competitive performance on both Composition-1k and the alphamatting.com benchmark quantitatively and qualitatively.

Dense Face Network:A Dense Face Detector Based on Global Context and Visual Attention Mechanism

Face detection has achieved tremendous strides thanks to convolutional neural networks. However, dense face detection remains an open challenge due to large face scale variation, tiny faces, and serious occlusion. This paper presents a robust, dense face detector using global context and visual attention mechanisms which can significantly improve detection accuracy. Specifically, a global context fusion module with top-down feedback is proposed to improve the ability to identify tiny faces. Moreover, a visual attention mechanism is employed to solve the problem of occlusion. Experimental results on the public face datasets WIDER FACE and FDDB demonstrate the effectiveness of the proposed method.

Improved Global Context Descriptor for Describing Interest Regions

The global context(GC) descriptor is improved for describing interest regions,uses gradient orientation for binning,and thus provides more robust invariance for geometric and photometric transformations.The performance of the improved GC(IGC) to image matching is studied through extensive experiments on the Oxford A?ne dataset.Empirical results indicate that the proposed IGC yields quite stable and robust results,signi?cantly outperforms the original GC,and also can outperform the classical scale-invariant feature transform(SIFT) in most of the test cases.By integrating the IGC to the SIFT,the resulting of hybrid SIFT+IGC performs best over all other single descriptors in these experimental evaluations with various geometric transformations.

Document-Level Neural Machine Translation with Hierarchical Modeling of Global Context

Document-level machine translation(MT)remains challenging due to its difficulty in efficiently using documentlevel global context for translation.In this paper,we propose a hierarchical model to learn the global context for documentlevel neural machine translation(NMT).This is done through a sentence encoder to capture intra-sentence dependencies and a document encoder to model document-level inter-sentence consistency and coherence.With this hierarchical architecture,we feedback the extracted document-level global context to each word in a top-down fashion to distinguish different translations of a word according to its specific surrounding context.Notably,we explore the effect of three popular attention functions during the information backward-distribution phase to take a deep look into the global context information distribution of our model.In addition,since large-scale in-domain document-level parallel corpora are usually unavailable,we use a two-step training strategy to take advantage of a large-scale corpus with out-of-domain parallel sentence pairs and a small-scale corpus with in-domain parallel document pairs to achieve the domain adaptability.Experimental results of our model on Chinese-English and English-German corpora significantly improve the Transformer baseline by 4.5 BLEU points on average which demonstrates the effectiveness of our proposed hierarchical model in document-level NMT.

The breakfast imperative： The changing context of global food security

The debate on global food security has regained vigor since the food crisis of 2008, when a sudden spike in the prices of staple food commodities dramatically demonstrated that securing the supply and accessibility of food for a world of nine billion people in 2050 cannot be taken for grant- ed （Godfray etal. 2010; Swinnen and Squicciarini 2012;

Specifying the Global Execution Context of Computer-Mediated Tasks: A Visual Notation and a Supporting Tool

This paper presents the notion of the global execution context of a task as a representational construct for analysing complexity in software evolution. Based on this notion a visual notation and a supporting tool are presented to support specification of a system’s global execution context. A system’s global execution context is conceived as an evolving network of use scenarios depicted by nodes and links designating semantic relationships between scenarios. A node represents either a base or a growth scenario. Directed links characterize the transition from one node to another by means of semantic scenario relationships. Each growth scenario is generated following a critique (or screening) of one or more base or reference scenarios. Subsequently, representative growth scenarios are compiled and consolidated in the global execution context graph. The paper describes the stages of this process, presents the tool designed to facilitate the construction of the global execution context graph and elaborates on recent practice and experience.

轻量级重参数化的遥感图像超分辨率重建网络设计

针对当前基于深度学习的遥感图像超分辨率重建模型部署时对硬件要求较高,本文设计了一种轻量级基于重参数化的残差特征遥感图像超分辨率重建网络。首先,采用重参数化方法设计了一种残差局部特征模块,以有效地提取图像局部特征;同时考虑到图像内部出现的相似特征,设计了一个轻量级的全局上下文模块对图像的相似特征进行关联以提升网络的特征表达能力,并通过调整该模块的通道压缩倍数来减少模型的参数量和改善模型的性能;最后,在上采样模块前使用多层特征融合模块聚合所有的深度特征,以产生更全面的特征表示。在UC Merced遥感数据集上进行测试,该算法在遥感图像3倍超分辨率下的参数量为539 K,峰值信噪比为30.01 dB,结构相似性为0.8449,模型的推理时间为0.010 s;而HSENet算法的参数量为5470 K,峰值信噪比为30.00 dB,结构相似性为0.8420,模型的推理时间为0.059 s。实验结果表明,该算法相比HSENet算法,参数量更少,运行速度较快,且峰值信噪比与结构相似性也有一定的提高。在DIV2K自然图像数据集上进行测试,该算法的峰值信噪比和结构相似性相比其他算法也有一定的优势,表明该算法的泛化能力较强。

The Trends of Globalization and Digitalization are Changing the Market Contexts

We know that SME’s that trade online grow faster and create more jobs than those that only operate in their domestic markets.The Internet is breaking down many traditional barriers to global trade,but there is still much governments can do to speed and enable SME digitization and ecommerce.The opportunity is huge at

双注意力随机选择全局上下文细粒度识别网络

针对细粒度图像识别任务中易忽视微小潜在性特征且外观差异细微等问题,提出一种基于双注意力随机选择全局上下文细粒度识别网络。首先,使用ConvNeXt作为主干网络,提出双注意力随机选择模块,对不同阶段提取到的特征进行通道随机选择和空间随机选择,使网络能够关注到其他潜在微小判别性特征;其次,利用全局上下文注意力模块将深层特征的语义信息融合到中间层,增强中间层定位微小特征的能力;最后,提出一种多分支损失,对中间层、深层和拼接层特征引入分类损失,结合不同分支提取到的特征,诱导网络获得多样性的判别特征。所提网络在Stanford-cars、CUB-200-2011、FGVC-Aircraft 3个公开细粒度数据集和真实场景下车型数据集VMRURS上分别达到了95.2%、92.1%、94.0%和97.0%的识别准确率,其性能相比其他对比方法有较大幅度提升。

基于多尺度上下文的英文作文自动评分研究

目前作文自动评分模型缺乏对不同尺度上下文语义特征的提取,未能从句子级别计算与作文主题关联程度的特征。提出基于多尺度上下文的英文作文自动评分研究方法MSC。采用XLNet英文预训练模型提取原始作文文本单词嵌入和句嵌入,避免在处理长序列文本时无法准确捕捉到符合上下文语境的向量嵌入,提升动态向量语义表征质量,解决一词多义问题,并通过一维卷积模块提取不同尺度的短语级别嵌入。多尺度上下文网络通过结合内置自注意力简单循环单元和全局注意力机制,分别捕捉单词、短语和句子级别的作文高维潜在上下文语义关联关系,利用句向量与作文主题计算语义相似度提取篇章主题层次特征,将所有特征输入融合层通过线性层得到自动评分结果。在公开的标准英文作文评分数据集ASAP上的实验结果表明,MSC模型平均二次加权的Kappa值达到了80.5%,且在多个子集上取得了最佳效果,优于实验对比的深度学习自动评分模型,证明了MSC在英文作文自动评分任务上的有效性。

基于全局与局部多尺度上下文的电表数据检测

电力系统中配电箱的电表数据检测为电力管理和安全运行提供了重要的数据支持。传统的人工电表数据读取方法效率低下且易出错,而现有深度学习方法因模型参数量大限制了模型的应用。针对上述问题,提出了一种轻量化鲁棒的实时电表检测方法。通过减少特征提取网络的层数和通道数,减少模型的参数量,实现网络的轻量化。在减少网络参数量的同时,为了保证网络的特征表达能力和拟合能力,引入全局上下文和局部多尺度上下文丰富目标特征表达。全局上下文关注电表数据在电表箱中的位置,局部多尺度上下文适应不同尺寸的电表数据。实验结果表明,所提网络在参数量更小的情况下,仍能获得比其他检测方法更高的准确率和更快的检测速度。

改进多头注意力机制的车道检测方法

针对基于卷积神经网络(convolution neural network,CNN)的车道线检测方法存在的网络处理效率低和对车道线细长结构的建模能力不佳的问题,提出一种基于改进多头注意力机制(multi-head self-attention,MHSA)的轻量级车道检测方法。引入MHSA,融合Fuse MBConv、MBConv模块与特征压缩模块,降低模型的参数,同时利用上下文信息嵌入模块,建立兼顾检测精度和推理速度的全局注意力网络;利用Transformer的编码和解码器以及前向反馈网络将车道线参数化,结合匈牙利拟合损失函数提高所提出方法对车道线细长结构的建模能力。在TuSimple数据集对所提出的方法进行验证,结果表明,所提出的方法识别精度达到96.3%,推理速度达到95帧/s,同时在Apollo无人驾驶平台上的运行速度达到60帧/s,能够满足实时检测的要求。

全局情境约束和局部多因素融合的对话情感识别

文本对话情感识别是自然语言处理领域中的一项重要研究任务,旨在自动识别对话文本各语句情感。然而,现有研究多侧重于对话语句的语义表征或对话人关系建模,忽略了对话交互过程中情感动态演变的影响因素。文章提出一种全局情境约束和局部多因素融合的对话文本情感识别方法,该模型不仅考虑了对话全局语义,还深入挖掘和建模了对话情感演变的多影响因素。在公开数据集上的实验结果表明该方法识别对话情感的有效性。

融合全局上下文注意力的遥感图像检测方法

针对遥感图像场景复杂、目标尺寸不一、且小尺寸目标过多导致的检测精度不佳和出现漏检等问题,提出了一种融合全局上下文注意力的目标检测算法。该算法提出一种全局上下文注意力机制和YOLOv5中C3结构融合的模块,以提升网络捕捉图像全局特征的能力;通过Varifocal Loss损失函数来提升对密集、尺寸小的目标的检测性能;采用基于归一化的注意力模块,降低图像中不太显著的特征和权重,使网络能够达到更高的检测准确率;利用动态卷积学习各个维度的信息,让训练得到的模型在降低GFLOPs情况下,同时保持检测精度提升。在NWPU VHR-10数据集上实验结果mAP为96.0%、准确率为98.2%、召回率为94.9%,较原YOLOv5模型分别提升了1.8%、4.7%和2.2%,证明了所改进YOLOv5方法的有效性。

一种弱纹理目标立体匹配网络

现有深度估计方法在高分辨率图像下存在特征提取不够充分、局部信息特征提取差的问题,为此提出一种面向全局特征的Transformer立体匹配网络;该网络采用编码器-解码器的端到端架构,采用多头注意力机制,允许模型在不同子空间中关注不同特征,从而提高特征提取能力;模型将自注意力机制和特征重构窗口结合,能够提高特征的表征能力,弥补局部特征不足的问题,在减少计算负担的同时有效解决Transformer架构计算复杂度高的问题,将模型的计算复杂度保持在线性范围内;在Scene Flow、KITTI-2015数据集上分别进行实验,与现有方法相比,相关指标得到显著提升,验证了模型的有效性和实用性。

正在书写的“文本”:全球人文视野下的国别与区域研究再思考

国别与区域研究为国内高校外语学科博士点五个主要方向之一,也是目前外语教学与研究的重点之一。究其根本,该学科源于西方“地理大探索”之际及随后的殖民扩张时代,兴于二战之后以美国为首的西方阵营对国际关系与战略竞争之筹措,在美、苏两大阵营对抗的冷战年代服务于各自全球的利益,故受到高度关注和重视。随着世界进入后冷战与全球化时代,这一学科不再仅仅局限于服从战略、政治、经济利益需求,而是更加深入拓展至语言、文化、哲学、艺术等人文领域,并进一步催生了后殖民研究、跨文化交流等相关学科之发展。本文拟就该学科之起源、发展、演变做一简要回顾,在当今全球人文语境/视野下予以再思考,探索外语学科国别与区域研究的发展途径、需要警惕的误区和朝向“语言文化”的国别与区域研究之可能性,以使其更好地服务于当今我国“一带一路”机制建设、铸牢中华民族共同体意识和中国式现代化之实践。