基于Class Balanced Loss修正交叉熵的非均衡样本信用风险评价模型

针对传统信用风险预测模型存在对非违约样本识别过度、对违约样本识别不足的问题,将平衡损失Class Balanced Loss函数引入信用风险评价,构建Class Balanced Loss修正交叉熵的非均衡样本信用风险评价模型。利用所建模型与交叉熵神经网络、支持向量机、决策树、随机森林和K最近邻5种分类模型进行对比,验证BPNN-CBCE对中国某金融机构1 534笔农户贷款数据信用风险预测的有效性;在此基础上,利用UCI公开的德国信贷数据验证BPNN-CBCE模型的稳健性。研究表明:对于农户数据,BPNN-CBCE模型在AUC、违约召回率Default recall方面普遍优于BPNN-CE、SVM、DT、RF和KNN模型,其中,BPNN-CBCE的Default recall相比5种对比模型提升了41.3个百分点,AUC相比5种对比模型提升了15.6个百分点;对于德国数据集,BPNN-CBCE评级模型在AUC、违约召回率Default recall方面也均优于5种对比模型。因此,BPNN-CBCE信用评价模型对农户不均衡信贷数据中的违约样本具有较好的识别能力,可有效降低金融机构客户误判带来的损失。创新与特色:①利用Class Balanced Loss中的平衡因子ω,增大违约样本在目标损失中的权重、降低非违约样本在目标损失中的权重,客观调节正负样本损失在目标损失中权重,弥补交叉熵函数无法调节两类样本损失权重的缺陷,克服由样本不均衡带来的评价模型对非违约样本识别过度、对违约样本识别不足。②通过考虑数据重叠,利用随机覆盖方法,分别对贷款数据中违约、非违约样本进行不放回采样,以对全样本空间X_(违约)、X_(非违约)进行不重叠覆盖,计算两类贷款客户的有效样本数量。既反映由于真实数据之间的内在相似性,随着样本数量的增加,新添加样本很可能是现有样本近似重复的客观事实,也保证基于有效样本对两类样本损失进行重新加权的客观性。将图像识别领域中的Class Balanced Loss函数引入信用评价领域,既拓展了Class Balanced Loss的使用边界,也为解决不均衡样本的信用风险评价提供了新的研究思路。

Balance Analysis of Profit and Loss on Exploiting Coal-Bed Gas in China

Models about four aspects according to the balance principle and practice in China were established, which involve the minimum production scale, alert production scale, safe running scale, and the goal production scale for specified profit level. It provides an effective quantitative analyzing method for the investors of coal bed gas exploitation project.

Strong Interaction Effect on Jet Energy Loss with Detailed Balance

The strong force effect on gluon distribution of quark-gluon plasma and its influence on jet energy loss with detailed balance are studied. We solve the possibility equation and obtain the value of non-extensive parameter q. In the presence of strong interaction, more gluons stay at low-energy state than the free gluon case. The strong interaction effect is found to be important for jet energy loss with detailed balance at intermediate jet energy. The energy gain via absorption increases with the strong interaction. This will affect the nuclear modification factor RAA and the parameter of q at intermediate jet energy.

Intelligent identification method and application of seismic faults based on a balanced classification network

This study combined fault identification with a deep learning algorithm and applied a convolutional neural network(CNN)design based on an improved balanced crossentropy(BCE)loss function to address the low accuracy in the intelligent identification of seismic faults and the slow training speed of convolutional neural networks caused by unbalanced training sample sets.The network structure and optimal hyperparameters were determined by extracting feature maps layer by layer and by analyzing the results of seismic feature extraction.The BCE loss function was used to add the parameter which is the ratio of nonfaults to the total sample sets,thereby changing the loss function to find the reference of the minimum weight parameter and adjusting the ratio of fault to nonfault data.The method overcame the unbalanced number of sample sets and improved the iteration speed.After a brief training,the accuracy could reach more than 95%,and gradient descent was evident.The proposed method was applied to fault identification in an oilfield area.The trained model can predict faults clearly,and the prediction results are basically consistent with an actual case,verifying the effectiveness and adaptability of the method.

Effect of potassium on soil conservation and productivity of maize/cowpea based crop rotations in the north-west Indian Himalayas

Plots under conservation tillage may require higher amount of potassium(K) application for augmenting productivity due to its stratification in upper soil layers, thereby reducing K supplying capacity in a medium or long-term period. To test this hypothesis, a field experiment was performed in 2002-2003 and 2006-2007 to study the effect of K and several crop rotations on yield, water productivity, carbon sequestration, grain quality, soil K status and economic benefits derived in maize(Zea mays L)/cowpea(Vigna sinensis L.) based cropping system under minimum tillage(MT). All crops recorded higher grain yield with a higher dose of K(120 kg K2 O ha-1) than recommended K(40 kg K2 O ha-1). The five years' average yield data showed that higher K application(120 kg K2 O ha-1) produced 16.4%(P<0.05)more maize equivalent yield. Cowpea based rotation yielded 14.2%(P<0.05) higher production than maize based rotation. The maximum enhancement was found in cowpea-mustard rotation. Relationship between yield and sustainable indices revealed that only agronomic efficiency of fertilizer input was significantly correlated with yield. Similarly, higherdoses of K application not only increased the water use efficiency(WUE) of all crops, but also reduced runoff and soil loss by 16.5% and 15.8% under maize and 23.3% and 19.7% under cowpea cover, respectively. This study also revealed that on an average 16.5% of left over carbon input contributed to soil organic carbon(SOC). Here, cowpea based rotation with the higher K application increased carbon sequestration in soil. Potassium fertilization also significantly improved the nutritional value of harvested grain by increasing the protein content for maize(by 9.5%) and cowpea(by 10.6%). The oil content in mustard increased by 5.0% and 6.0% after maize and cowpea, respectively. Net return also increased with the application of the higher K than recommended K and the trend was similar to yield. Hence, the present study demonstrated the potential yield and profit gains along with resource conservation in the Indian Himalayas due to annual additions of higher amount of K than the recommended dose. The impact of high K application was maximum in the cowpea-mustard rotation.

General Admissibility for Linear Estimators in a General Multivariate Linear Model under Balanced Loss Function

A generalization of Zellner’s balanced loss function is proposed. General admissibility in a general multivariate linear model is investigated under the generalized balanced loss function. And the sufficient and necessary conditions for linear estimators to be generally admissible in classes of homogeneous and nonhomogeneous linear estimators are given, respectively.

B-PesNet: Smoothly Propagating Semantics for Robust and Reliable Multi-Scale Object Detection for Secure Systems

Multi-scale object detection is a research hotspot,and it has critical applications in many secure systems.Although the object detection algorithms have constantly been progressing recently,how to perform highly accurate and reliable multi-class object detection is still a challenging task due to the influence of many factors,such as the deformation and occlusion of the object in the actual scene.The more interference factors,the more complicated the semantic information,so we need a deeper network to extract deep information.However,deep neural networks often suffer from network degradation.To prevent the occurrence of degradation on deep neural networks,we put forth a new model using a newly-designed Pre-ReLU,which inserts a ReLU layer before the convolution layer for the sake of preventing network degradation and ensuring the performance of deep networks.This structure can transfer the semantic information more smoothly from the shallow to the deep layer.However,the deep networks will encounter not only degradation,but also a decline in efficiency.Therefore,to speed up the two-stage detector,we divide the feature map into many groups so as to diminish the number of parameters.Correspondingly,calculation speed has been enhanced,achieving a balance between speed and accuracy.Through mathematical demonstration,a Balanced Loss(BL)is proposed by a balance factor to decrease the weight of the negative sample during the training phase to balance the positives and negatives.Finally,our detector demonstrates rosy results in a range of experiments and gains an mAP of 73.38 on PASCAL VOC2007,which approaches the requirement of many security systems.

Fate of Basal N Under Split Fertilization in Rice with ^(15)N Isotope Tracer

Split fertilization strategy is popularly adopted in rice to synchronize soil nitrogen(N) supply and crop N demand. Attention has been paid more on mid-season topdressing N, but limited on basal N. A clearer understanding of the basal N fate under split fertilization is crucial for determining rational basal N split ratio to improve the yield and reduce the loss to environment. A two-year field experiment with two N rates of 150 and 300 kg Nha^(-1), two split ratios of basal N, 40% and 25%, and two rice varieties,Wuyunjing 23(japonica) and Y-liangyou 2(super hybrid indica), was conducted. Labelled ^(15) N urea was supplied in micro-plots as basal fertilizer to determine the plant uptake, translocation, soil residual, and loss of basal N fertilizer. The results showed that basal N absorbed by rice was only 1.6%–11.5% before tillering fertilization(8–10 d after transplanting), 6.5%–21.4% from tillering fertilization to panicle fertilization, and little(0.1%–4.4%) after panicle fertilization. The recovery efficiency of basal N for the entire rice growth stage was low and ranged from 18.7% to 24.8%, not significantly affected by cultivars or N treatments. Soil residual basal N accounted for 10.3%–36.4% and decreased with increasing total N rate and basal N ratio, regardless of variety and year. 43.8%–70.4% of basal N was lost into the environment based on the N balance. Basal N loss was significantly linearly positive related with the basal N rate and obviously enhanced by the increasing basal N ratio for both varieties in both 2012 and 2013. The N use efficiency and yield was significantly improved when decreasing the basal N ratio from 40% to 25%. The results indicated that the basal N ratio should be reduced, especially with limited N inputs, to improve the yield and reduce the N loss to the environment.

A novel circuit architecture for fourth subharmonic mixers

A circuit topology for high-order subharmonic（SH） mixers is described.By phase cancellation of idle frequency components,the SH mixer circuit can eliminate the complicated design procedure of idle frequency circuits.Similarly,the SH mixer circuit can achieve a high port isolation by phase cancellation of the leakage LO, RF and idle frequency signals.Based on the high-order SH mixer architecture,a new Ka-band fourth SH mixer is analyzed and designed,it shows the lowest measured conversion loss of 8.3 dB at 38.4 GHz and the loss is lower than 10.3 dB in 34-39 GHz.Measured LO-IF,RF-LO,RF-IF port isolation are better than 30.7 dB,22.9dB and 46.5 dB,respectively.