XGBoost Algorithm under Differential Privacy Protection

Privacy protection is a hot research topic in information security field.An improved XGBoost algorithm is proposed to protect the privacy in classification tasks.By combining with differential privacy protection,the XGBoost can improve the classification accuracy while protecting privacy information.When using CART regression tree to build a single decision tree,noise is added according to Laplace mechanism.Compared with random forest algorithm,this algorithm can reduce computation cost and prevent overfitting to a certain extent.The experimental results show that the proposed algorithm is more effective than other traditional algorithms while protecting the privacy information in training data.

Mechanical Eye Trauma Epidemiology, Prognostic Factors, and Management Controversies—An Update

Purpose of Review: The management of eye injuries is both difficult and argumentative. This study attempts to highlight the management of ocular trauma using currently available information in the literature and author experience. This review presents a workable framework from the first presentation, epidemiology, classification, investigations, management principles, complications, prognostic factors, final visual outcome and management debates. Review Findings: Mechanical ocular trauma is a leading cause of monocular blindness and possible handicap worldwide. Among several classification systems, the most widely accepted is Birmingham Eye Trauma Terminology (BETT). Mechanical ocular trauma is a topic of unsolved controversy. Patching for corneal abrasion, paracentesis for hyphema, the timing of cataract surgery and intraocular lens implantation are all issues in anterior segment injuries. Regarding posterior segment controversies, the timing of vitrectomy, use of prophylactic cryotherapy, the necessity of intravitreal antibiotics in the absence of infection, the use of vitrectomy vs vitreous tap in traumatic endophthalmitis is the issues. The pediatric age group needs to be approached by a different protocol due to the risk of amblyopia, intraocular inflammation, and significant vitreoretinal adhesions. The various prognostic factors have a role in the final visual outcome. B scan is used to exclude R.D, Intraocular foreign body (IOFB), and vitreous haemorrhage in hazy media. Individual surgical strategies are used for every patient according to the classification and extent of the injuries. Conclusion: This article examines relevant evidence on the management challenges and controversies of mechanical trauma of the eye and offers treatment recommendations based on published research and the authors’ own experience.

北京市不同功能区不透水地表时空变化差异分析（英文）

Impervious surface(IS) is often recognized as the indicator of urban environmental changes. Numerous research efforts have been devoted to studying its spatio-temporal dynamics and ecological effects, especially for the IS in Beijing metropolitan region. However, most previous studies primarily considered the Beijing metropolitan region as a whole without considering the differences and heterogeneity among the function zones. In this study, the subpixel impervious surface results in Beijing within a time series(1991, 2001, 2005, 2011 and 2015) were extracted by means of the classification and regression tree(CART) model combined with change detection models. Then based on the method of standard deviation ellipse, Lorenz curve, contribution index(CI) and landscape metrics, the spatio-temporal dynamics and variations of IS(1991, 2001, 2011 and 2015) in different function zones and districts were analyzed. It is found that the total area of impervious surface in Beijing increased dramatically during the study period, increasing about 144.18%. The deflection angle of major axis of standard deviation ellipse decreased from 47.15° to 38.82°, indicating the major development axis in Beijing gradually moved from northeast-southwest to north-south. Moreover, the heterogeneity of impervious surface’s distribution among 16 districts weakened gradually, but the CI values and landscape metrics in four function zones differed greatly. The urban function extended zone(UFEZ), the main source of the growth of IS in Beijing, had the highest CI values. Its lowest CI value was 1.79 that is still much higher than the highest CI value in other function zones. The core function zone(CFZ), the traditional aggregation zone of impervious surface, had the highest contagion index(CONTAG) values, but it contributed less than UFEZ due to its small area. The CI value of the new urban developed zone(NUDZ) increased rapidly, and it increased from negative to positive and multiplied, becoming animportant contributor to the rise of urban impervious surface. However, the ecological conservation zone(ECZ) had a constant negative contribution all the time, and its CI value decreased gradually. Moreover, the landscape metrics and centroids of impervious surface in different density classes differed greatly. The high-density impervious surface had a more compact configuration and a greater impact on the eco-environment.

A Machine Learning Approach for Collusion Detection in Electricity Markets Based on Nash Equilibrium Theory

We aim to provide a tool for independent system operators to detect the collusion and identify the colluding firms by using day-ahead data. In this paper, an approach based on supervised machine learning is presented for collusion detection in electricity markets. The possible scenarios of the collusion among generation firms are firstly identified. Then,for each scenario and possible load demand, market equilibrium is computed. Market equilibrium points under different collusions and their peripheral points are used to train the collusion detection machine using supervised learning approaches such as classification and regression tree(CART) and support vector machine(SVM) algorithms. By applying the proposed approach to a four-firm and ten-generator test system, the accuracy of the proposed approach is evaluated and the efficiency of SVM and CART algorithms in collusion detection are compared with other supervised learning and statistical techniques.