Incomplete Block Designs (Lecture Notes in Statistics,Vol. l)

Because mixtures of exponential distributions (with constant failure rate) have the decreasing failure rate (DFR) property, as shown by Proschan in the sixties, it is not unexpected that mixtures of distributions that have a mildly increasing failure rate (IFR) also have this property. What is. perhaps, surprising is that mixtures of IFR distributions with rapidly increasing failure rate may also behave in this manner. Two striking examples of such mixtures are presented here and illustrated graphically. This phenomenon could be the cause of concern, in practice, when it is suspected that a sample is based on pooled data.     

LRHW-AP: Using ranking-based metric as loss for Person Re-Identification

Optimizing a ranking-based metric as the loss function, such as Average Precision (AP), has been found very effective in image retrieval tasks, but it has received less attention in Person Re-Identification (Re-ID). In this paper, Low Rank High Weight (LRHW) AP is proposed to apply the AP-optimizing method on the Re-ID task. LRHW-AP employs high weight on the low rank positive instances, which provides more information for model optimization than high rank positive instances and distribute in high gradient area. We propose a new pooling method called Power Activation Weighted Mean (PAWM) pooling which can unify a set of pooling methods because of a changeable activation function and a trainable parameter. Thus one can adjust and train PAWM to adapt to the target task to improve the model performance. Besides, we incorporate Warmup and Exponentially Decay Scheduler with a delay period, called Warmup Delay Exponentially Decay Scheduler, which brings further improvement. Through an extensive set of ablation studies, we verify that all methods mentioned above contribute to the performance boosts on Re-ID and the model achieves 95.3% rank-1 and 88.4% mAP on Market1501 with ResNet50.     

基于轻量级卷积神经网络的植物叶片病害识别方法

针对目前农业信息领域植物病害识别精度较低、实时性较差的问题,提出了一种基于轻量级卷积神经网络（CNN）的植物叶片病害识别方法。在原有网络中引入深度可分离卷积（DSC）和全局平均池化（GAP）方法,分别用来代替标准卷积运算操作并对网络末端的全连接层部分进行替换。同时,批归一化的技巧也被运用到训练网络的过程中,以改善中间层数据分布并提高收敛速度。为全面而可靠地评估所提方法的性能,在公开的植物叶片病害图像数据集PlantVillage上进行实验,选取损失函数收敛曲线、测试精度、参数内存需求等指标来验证改进策略的有效性。实验结果表明,改进后的网络具有较高的病害识别精度（99.427%）以及较小的内存空间占用（6.47 MB）,可见其与其他基于神经网络的叶片识别技术相比具有优势,工程实用性较强。     

基于改进的Faster R-CNN的通用目标检测框架

针对当前基于深度学习的检测器不能有效检测形状不规则或长宽相差悬殊的目标的问题,在传统Faster R-CNN算法的基础上,提出了一个改进的二阶段目标检测框架——Accurate R-CNN。首先,提出了新的交并比（IoU）度量——有效交并比（EIoU）,通过提出中心度权重来降低训练数据中冗余包围框的占比。然后,提出了一个上下文相关的特征重分配模块（FRM）,通过建模目标的远程依赖和局部上下文关系信息对特征进行重编码,以弥补池化过程中的形状信息损失。实验结果表明,在微软多场景通用目标（MS COCO）数据集上,对于包围框检测任务,当使用深度为50和101的残差网络（ResNet）作为骨干网络时,Accurate R-CNN比基线模型Faster R-CNN的平均精度（AP）分别提高了1.7个百分点和1.1个百分点,超越了使用同样骨干网络的基于掩膜的检测器。在添加掩膜分支后,对于实例分割任务,当使用两种不同深度的ResNet作为骨干网络时,Accurate R-CNN比Mask R-CNN的掩膜平均精度分别提高了1.2个百分点和1.1个百分点。研究结果显示,相较于基线模型,Accurate R-CNN在不同数据集、不同任务上均取得了更好的检测效果。     

SGSN＋MME融合组Pool策略研究

基于3GPP核心网融合演进背景和池组化技术分析,探讨了SGSN与MME融合组Pool的技术挑战以及相应解决方案,其中包括网络覆盖区域规划和网间标识映射方案,并对上述技术方案进行比较,最后针对现网部署给出网络融合组Pool策略的建议。     

基于深度卷积神经网络的轴承故障诊断技术研究

为了提高轴承故障诊断的准确度，采用深度卷积神经网络算法来实现轴承故障分类。首先根据轴承振动故障特征频率建立轴承故障数据库，接着对轴承的振动信号按不同切片长度和固定宽度进行周期提取，建立特征向量矩阵，然后建立深度卷积神经网络的故障诊断模型，在网络设计时，差异化设置卷积核与池化尺寸，优化神经网络训练的核心参数，最后获得稳定的卷积神经网络模型。经过实例仿真，基于深度卷积神经网络的轴承故障分类准确率高，标准差小。     

多层次深度网络融合人脸识别算法*

深度学习模型可以获得更具有鉴别力的人脸特征,提高人脸识别性能.因此,文中结合深度学习思想,提出多层次深度网络融合特征提取模型.在深度子空间基础上,采用“卷积-池化”网络结构,在降低特征维度的同时保留图像纹理信息,并且获得局部转换鲁棒性.同时,利用人脸标定算法获得人脸特征点,并以此划分人脸区域为5个局部人脸块.基于多层次分类策略,利用全局人脸训练全局网络,完成测试样本预分类.利用局部人脸块训练局部网络,在候选类别中完成最终分类.实验表明,结合局部特征与全局特征的模型可以取得较好的识别率,对光照、表情、姿态,遮挡等影响因素具有较好的鲁棒性,并且加入池化层及两步判别的算法可以有效提高识别率.     

融合局部性和非负性的Laplacian稀疏编码的图像分类*

稀疏编码在编码过程中忽略特征之间的局部关系,使编码不稳定,并且优化问题中的减法运算可能会导致特征之间相互抵消.针对上述2个问题,文中提出融合局部性和非负性的Laplacian稀疏编码的图像分类方法.引入局部特征附近的基约束编码,利用非负矩阵分解将非负性加到Laplacian稀疏编码中,利用空间金字塔划分和最大值融合表示最终的图像,并采用多类线性SVM分类图像.本文方法保留特征之间的局部信息,避免特征之间相互抵消,保留更多的特征,从而改善编码的不稳定性.在4个公共数据集上的实验表明,相比其它现有算法,本文方法分类准确率更高.     

改进SegNet+CRF高分辨率遥感影像建筑物提取方法

将传统的语义分割SegNet网络用于高分辨率遥感影像的建筑物提取时,分割的建筑物存在边界模糊、精度较低、错检漏检等问题。为了解决上述问题,提出一种改进SegNet网络+CRF语义分割方法。编码阶段的最低分辨率层引入空洞金字塔池化模型,通过并行的空洞卷积操作扩大特征提取的感受野;解码阶段构建特征金字塔实现特征多尺度融合,弥补上采样过程中丢失的特征信息;最后,预测图像送入全连接条件随机场模型进行后处理,优化提取的建筑物边缘。实验表明,相较于原SegNet网络,改进方法的建筑物提取像素精度、召回率、平均交并比分别提高了0.48%、1.29%、2.36%。     

SKU demand forecasting in the presence of promotions

Promotions and shorter life cycles make grocery sales forecasting more difficult, requiring more complicated models. We identify methods of increasing complexity and data preparation cost yielding increasing improvements in forecasting accuracy, by varying the forecasting technique, the input features and model scope on an extensive SKU-store level sales and promotion time series from a European grocery retailer. At the high end of data and technique complexity, we propose using regression trees with explicit features constructed from sales and promotion time series of the focal and related SKU-store combinations. We observe that data pooling almost always improves model performance. The results indicate that simple time series techniques perform very well for periods without promotions. However, for periods with promotions, regression trees with explicit features improve accuracy substantially. More sophisticated input is only beneficial when advanced techniques are used. We believe that our approach and findings shed light into certain questions that arise while building a grocery sales forecasting system.