Inverse reinforcement learning-based time-dependent A* planner for human-aware robot navigation with local vision

In an environment where robots coexist with humans, mobile robots should be human-aware and comply with humans' behavioural norms so as to not disturb humans' personal space and activities. In this work, we propose an inverse reinforcement learning-based time-dependent A* planner for human-aware robot navigation with local vision. In this method, the planning process of time-dependent A* is regarded as a Markov decision process and the cost function of the time-dependent A* is learned using the inverse reinforcement learning via capturing humans' demonstration trajectories. With this method, a robot can plan a path that complies with humans' behaviour patterns and the robot's kinematics. When constructing feature vectors of the cost function, considering the local vision characteristics, we propose a visual coverage feature for enabling robots to learn from how humans move in a limited visual field. The effectiveness of the proposed method has been validated by experiments in real-world scenarios: using this approach robots can effectively mimic human motion patterns when avoiding pedestrians; furthermore, in a limited visual field, robots can learn to choose a path that enables them to have the larger visual coverage which shows a better navigation performance.     

命名实体识别技术综述

命名实体识别是自然语言处理中的热点研究方向之一,目的是识别文本中的命名实体并将其归纳到相应的实体类型中。首先阐述了命名实体识别任务的定义、目标和意义,分析提出了命名实体识别的主要难点在于领域命名实体识别局限性、命名实体表述多样性和歧义性、命名实体的复杂性和开放性;然后介绍了命名实体识别研究的发展进程,从最初的规则和字典方法到传统的统计学习方法再到现在的深度学习方法,不断地将新技术应用到命名实体识别研究中以提高性能;接着系统梳理了当下命名实体识别任务中的若干热门研究点,分别是匮乏资源下的命名实体识别、细粒度命名实体识别、嵌套命名实体识别以及命名实体链接;最后针对评判命名实体识别模型的好坏,总结了常用的若干数据集和实验测评指标,并给出了未来的研究建议。     

伪标签置信选择的半监督集成学习视频语义检测

在视频语义检测中，有标记样本不足会严重影响检测的性能，而且伪标签样本中的噪声也会导致集成学习基分类器性能提升不足。为此，提出一种伪标签置信选择的半监督集成学习算法。首先，在三个不同的特征空间上训练出三个基分类器，得到基分类器的标签矢量；然后，引入加权融合样本所属某个类别的最大概率与次大概率的误差和样本所属某个类别的最大概率与样本所属其他各类别的平均概率的误差，作为基分类器的标签置信度，并融合标签矢量和标签置信度得到样本的伪标签和集成置信度；接着，选择集成置信度高的样本加入到有标签的样本集，迭代训练基分类器；最后，采用训练好的基分类器集成协作检测视频语义概念。该算法在实验数据集UCF11上的平均准确率到达了83.48%，与Co-KNN-SVM算法相比，平均准确率提高了3.48个百分点。该算法选择的伪标签能体现样本所属类别与其他类别的总体差异性，又能体现所属类别的唯一性，可减少利用伪标签样本的风险，有效提高视频语义概念检测的准确率。     

改进的卷积神经网络单幅图像超分辨率重建

针对经典的基于卷积神经网络的单幅图像超分辨率重建方法网络较浅、提取的特征少、重建图像模糊等问题，提出了一种改进的卷积神经网络的单幅图像超分辨率重建方法，设计了由密集残差网络和反卷积网络组成的新型深度卷积神经网络结构。原始低分辨率图像输入网络，利用密集残差学习网络获取更丰富的有效特征并加快特征梯度流动，其次通过反卷积层将图像特征上采样到目标图像大小，再利用密集残差学习高维特征，最后融合不同卷积核提取的特征得到最终的重建图像。在Set5和Set14数据集上进行了实验，并和Bicubic、K-SVD、SelfEx、SRCNN等经典重建方法进行了对比，重建出的图像在整体清晰度和边缘锐度方面更好，另外峰值信噪比（PSNR）平均分别提高了2.69?dB、1.68?dB、0.74?dB和0.61?dB。实验结果表明，该方法能够获取更丰富的细节信息，得到更好的视觉效果，达到了图像超分辨率的增强任务。     

Convolutional neural network: Deep learning-based classification of building quality problems

The rapid development of the construction industry in China has introduced unprecedented quality-related problems in the country’s building industry. In response to this issue, the government has established various complaint channels to report quality problems. Therefore, building quality complaints (BQCs) need to be classified and solved by respective agencies or departments rapidly for avoiding adverse impact on the safety, health, and well-being of people. However, the current process of classifying BQCs is labor intensive, time consuming, and error prone. An automatic complaint classification is required to improve the effectiveness and efficiency of complaint handling, but studies on this issue are limited. Prevailing text classification research in construction has focused on utilizing conventional shallow machine learning. By contrast, this study explores a novel convolutional neural network (CNN)-based approach that incorporates a deep-learning method to automatically classify the short texts contained within BQCs. The presented approach enables capturing the semantic features in BQC texts and automatic classification of the BQCs into predefined categories. After the model optimization, tests are conducted to examine the practical application of the text classification approach compared with Bayes-based and support vector machine classifiers. Results indicate that the developed CNN-based approach performs well in the Chinese BQC classification with limited manual intervention and few complicated feature engineering.     

基于优势学习的深度Q网络

强化学习问题中，同一状态下不同动作所对应的状态-动作值存在差距过小的现象，Q-Learning算法采用MAX进行动作选择时会出现过估计问题，且结合了Q-Learning的深度Q网络（Deep Q Net）同样存在过估计问题。为了缓解深度Q网络中存在的过估计问题，提出一种基于优势学习的深度Q网络，通过优势学习的方法构造一个更正项，利用目标值网络对更正项进行建模，同时与深度Q网络的评估函数进行求和作为新的评估函数。当选择的动作是最优动作时，更正项为零，不对评估函数的值进行改动，当选择的动作不是最优动作时，更正项的值为负，降低了非最优动作的评估值。和传统的深度Q网络相比，基于优势学习的深度Q网络在Playing Atari 2600的控制问题breakout、seaquest、phoenix、amidar中取得了更高的平均奖赏值，在krull、seaquest中取得了更加稳定的策略。     

结合深度学习与支持向量机的金属零件识别

目的 在视觉引导的工业机器人自动拾取研究中，关键技术难点之一是机器人抓取目标区域的识别问题。特别是金属零件，其表面的反光、随意摆放时相互遮挡等非结构化因素都给抓取区域的识别带来巨大的挑战。因此，本文提出一种结合深度学习和支持向量机的抓取区域识别方法。方法 分别提取抓取区域的方向梯度直方图（HOG）和局部二进制模式（LBP）特征，利用主成分分析法（PCA）对融合后的特征进行降维，以此来训练支持向量机（SVM）分类器。通过训练Mask R-CNN（regions with convolutional neural network）神经网络完成抓取区域的初步分割。然后利用SVM对Mask R-CNN识别的抓取区域进行二次分类，完成对干扰区域的剔除。最后计算掩码完成实例分割，以此达到对抓取区域的精确识别。结果 对于随机摆放的铜质金属零件，本文算法与单一的Mask R-CNN及多特征融合的SVM算法就识别准确率、错检率、漏检率3个指标进行了比较，结果表明本文算法在识别准确率上较Mask R-CNN和SVM算法分别提高了7%和25%，同时有效降低了错检率与漏检率。结论 本文算法结合了Mask R-CNN与SVM两种方法，对于反光和遮挡情况具有一定的鲁棒性，同时有效地提升了目标识别的准确率。     

Multiple Empirical Kernel Learning with Majority Projection for imbalanced problems

Traditional Multiple Empirical Kernel Learning (MEKL) expands the expressions of the sample and brings better classification ability by using different empirical kernels to map the original data space into multiple kernel spaces. To make MEKL suit for the imbalanced problems, this paper introduces a weight matrix and a regularization term into MEKL. The weight matrix assigns high misclassification cost to the minority samples to balanced misclassification cost between minority and majority class. The regularization term named Majority Projection (MP) is used to make the classification hyperplane fit the distribution shape of majority samples and enlarge the between-class distance of minority and majority class. The contributions of this work are: (i) assigning high cost to minority samples to deal with imbalanced problems, (ii) introducing a new regularization term to concern the property of data distribution, (iii) and modifying the original PAC-Bayes bound to test the error upper bound of MEKL-MP. Through analyzing the experimental results, the proposed MEKL-MP is well suited to the imbalanced problems and has lower generalization risk in accordance with the value of PAC-Bayes bound.     

Dialogic weaving: a favorable tension between design and craft

Focusing on a process where a designer embeds herself in craftspeople’s workshops with the intention of ‘learning from inside’ and ‘making together’; this paper proposes pathways for considering collaboration between designers and craftspeople via digital fabrication and weaving. The tension of forces that make weaving possible, act as an analogy for this collaboration as well as providing the title of the project as ‘One Over, One Under’. The project proposes a mode of working and a spectrum of outcomes where the designer has a first-hand experience of production techniques, engages in a serious and continuous dialogue through making, and develops an intervention that also brings forth her own skill set (introducing digital fabrication and structural variations), thus transforming the conventions and boundaries between established roles and manufacturing techniques. An experiment that resulted in a series of objects considering both the technologies of production and the input of the designer, this process not only increases the potentials that crafts hold for the field of design, but it also offers possibilities of collaboration and a further articulation of the design act.