深度学习识别光网络单元故障的设计与应用

为解决依赖装维上门鉴别光网络单元故障带来的不便,可以从机器视觉入手实现自动化故障识别。近年,ImageNet挑战赛的成功推动了物体识别技术的跨越式发展,特别是基于卷积的深度学习技术在视觉识别方面已经达到人类水平,为光网络单元故障的自动识别提供了技术基础。文章对识别光网络单元的工作状态进行了研究,将设备工作状态分为7个场景,提出了利用手机APP采集图片识别故障的解决方案并投入了实际生产;重点阐述了深度学习模块的设计与实现,提出一种通过算法整合的方式综合运用物体检测和图像分类算法,分3阶段逐步求精,解决了图片过滤,光网络单元型号和状态识别等问题,实现了基于计算机视觉自动识别光网络单元故障。从数据上看产品的端到端准确率超过84%,识别速度达到10 FPS,月均提供服务超过1万人次,在减少用户等待的同时节约了人力资源。     

Image and Feature Space Based Domain Adaptation for Vehicle Detection

The application of deep learning in the field of object detection has experienced much progress. However, due to the domain shift problem, applying an off-the-shelf detector to another domain leads to a significant performance drop. A large number of ground truth labels are required when using another domain to train models, demanding a large amount of human and financial resources. In order to avoid excessive resource requirements and performance drop caused by domain shift, this paper proposes a new domain adaptive approach to cross-domain vehicle detection. Our approach improves the cross-domain vehicle detection model from image space and feature space. We employ objectives of the generative adversarial network and cycle consistency loss for image style transfer in image space. For feature space, we align feature distributions between the source domain and the target domain to improve the detection accuracy. Experiments are carried out using the method with two different datasets, proving that this technique effectively improves the accuracy of vehicle detection in the target domain.     

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.     

基于特征融合的中文简历解析方法研究

针对基于规则和统计的传统中文简历解析方法效率低、成本高、泛化能力差的缺点，提出一种基于特征融合的中文简历解析方法，即级联Word2Vec生成的词向量和用BLSTM（Bidirectional Long Short-Term Memory）建模字序列生成的词向量，然后再结合BLSTM和CRF（Conditional Random Fields）对中文简历进行解析（BLSTM-CRF）。为了提高中文简历解析的效率，级联包含字序列信息的词向量和用Word2Vec生成的词向量，融合成一个新的词向量表示；再由BLSTM强大的学习能力融合词的上下文信息，输出所有可能标签序列的分值给CRF层；再由CRF引入标签之间约束关系求解最优序列。利用梯度下降算法训练神经网络，使用预先训练的词向量和Dropout优化神经网络，最终完成对中文简历的解析工作。实验结果表明，所提的特征融合方法优于传统的简历解析方法。