面向隐私保护联邦学习的医学影像目标检测算法

联邦学习技术是一种新型多机构协同训练模型范式,广泛应用于多领域,其中模型参数隐私保护是一个关键问题.针对CT影像综合性病灶检测任务,提出隐私保护的联邦学习算法.首先部署松散耦合的客户端-服务器架构;其次在各客户端使用改进的RetinaNet检测器,引入上下文卷积和后向注意力机制;最后完成联邦训练.各客户端使用局部更新策略,采用自适应训练周期,局部目标函数中加入了限制项;服务器使用自适应梯度裁剪策略和高斯噪声差分隐私算法更新全局模型参数.在DeepLesion数据集上的消融分析说明了算法各部分的重要性.实验结果表明,改进的RetinaNet检测器有效地提升了多尺度病灶的检测精度.与集中数据训练模型范式相比,联邦学习所得模型性能略低(mAP分别为75.33％和72.80％),但训练用时缩短近38％,有效地实现了隐私保护、通信效率和模型性能的良好权衡.     

RLINK: Deep reinforcement learning for user identity linkage

World Wide Web - User identity linkage is a task of recognizing the identities of the same user across different social networks (SN). Previous works tackle this problem via estimating the pairwise...     

基于稳健设计的SGNS算法的超参数调优方法

通常给定超参数的若干取值选取性能最大的为最优组合(称为直接选优法),但是此方法的稳健性差。为此,提出了一种基于稳健设计的超参数调优方法(称为稳健调优法)。具体地,以SGNS算法中的超参数调优为例,在词语推断任务上实验并得出:经方差分析得到SGNS算法中的七个超参数中的五个对算法预测性能有显著影响,确定为主控因子,其余两个确定为噪声因子,且主控因子中有三个对性能估计的方差有显著影响,因此,调优中仅从期望最大来直接选优是不合理的;稳健调优法与直接选优法两者在预测性能上没有显著差异,但稳健调优法对噪声因子具有较好的稳健性。稳健调优法对一般的深度神经网络的调参有实际的借鉴意义。     

Electrochemical properties of some cobalt antimonides as anode materi-als for lithium- ion batteries

Some cobalt antimonides have been prepared and studied as the candidate anode materials for lithium ion batter-ies. Reversible capacities of 424,423 and 546 mA·h·g-1 were measured at the first cycle for as-solidified CoSb2, CoSb3 and annealed CoSb3 respectively. A low lithium ions diffusion coefficient in the order of 10-16 m2·s-1 was estimated from the coulometric titration measurements in the annealed CoSb3 electrode. It was found that the electrochemical properties of fine powders are significantly better than coarse powders. However the SEM picture shows that the nano-sized CoSb3 powders gathered to larger granules, which worsens somewhat the capacity retention of the nano-sized materials, although the volume capacities of the annealed and ball milled CoSb3 remain near twice of that of graphite after 50 cycles.     

Preparation and characterization of GaN films grown on Ga-diffused Si （111） substrates

Hexagonal GaN films were prepared by nitriding Ga2O3 films with flowing ammonia. Ga2O3 films were deposited on Ga-diffused Si (111) substrates by radio frequency (r.f.) magnetron sputtering. This paper have investigated the change of structural properties of GaN films nitrided in NH3 atmosphere at the temperatures of 850, 900, and 950℃ for 15 min and nitrided at the temperature of 900℃ for 10, 15, and 20 rain, respectively. X-ray diffraction (XRD), scanning electron microscopy (SEM), transmission electron microscopy (TEM) and X-ray photoelectron spectroscopy (XPS) were used to analyze the structure, surface morphology and composition of synthesized samples. The results reveal that the as-grown films are polycrystalline GaN with hexagonal wurtzite structure and GaN films with the highest crystal quality can be obtained when nitrided at 900℃ for 15 min.     

钼磷系缓蚀阻垢剂SK-107的缓蚀、阻垢性能研究

通过大量缓蚀、阻垢实验,着重介绍了钼磷系缓蚀阻垢剂SK-107在高碱度水且局部漏氨的合成氨循环水系统中的缓蚀、阻垢性能。     

新型冷轧(拔)无缝钢管高分子润滑工艺研究

提出了冷轧(拔)无缝钢管高分子润滑工艺。研究了该工艺的PL膜各种特性。介绍了工业实验情况。实验证明,高分子润滑工艺完全满足冷轧(拔)钢管工艺要求,并具有操作简便、环境污染小等优点,完全可以代替传统的磷化-皂化润滑工艺。     

Compact difference method for solving the fractional reaction–subdiffusion equation with Neumann boundary value condition

In this paper, we derive a high-order compact finite difference scheme for solving the reaction–subdiffusion equation with Neumann boundary value condition. The L1 method is used to approximate the temporal Caputo derivative, and the compact difference operator is applied for spatial discretization. We prove that the compact finite difference method is unconditionally stable and convergent with order O(τ^2?α+h⁴) in L₂ norm, where τ, α, and h are the temporal step size, the order of time fractional derivative and the spatial step size, respectively. Finally, some numerical experiments are carried out to show the effectiveness of the proposed difference scheme.     

移动定位技术的现状与发展趋势

基于位置服务为人们的日常生活带来了极大的方便,同时也推动了移动定位技术行业的快速发展。针对移动定位技术发展状况,分析目前主流定位技术的优劣,并对取得的突破作简单的总结归纳。