分布式随机方差消减梯度下降算法topkSVRG

机器学习问题通常会转换成一个目标函数进行求解,优化算法是求解目标函数中参数的重要工具。随机梯度下降(stochastic gradient descent,SGD)是目前应用最广的算法,因其易受噪声干扰只能达到次线性收敛率,而改进后的随机方差消减梯度法(stochastic variance reduction gradient,SVRG)则可以达到线性的收敛率。SVRG是一种串行单机版算法,为了应对大规模数据集分布式训练问题,设计一种以SVRG算法思想为基础的分布式SVRG的实现算法topk SVRG。改进在于:主节点维护一个全局模型,从节点基于本地数据进行局部模型更新。每轮迭代时,选择与当前全局模型距离最小的k个局部模型进行平均来更新全局模型,参数k调大可以提高收敛速度,调小k可以保证收敛。理论分析了算法的线性收敛性,基于Spark进行算法实现,通过与Mini-Batch SGD、CoCoA、Splash及相关算法的实验比较,topkSVRG可以在高精度要求下更快地收敛。

Expression of human hepatitis C virus core antigen in tobacco plants by tobacco mosaic virus-based vector system

Tobacco mosaic virus (TMV) has the potential to highly express foreign gene. A novel TMV-based in trans expression system was constructed. A TMV mutant TSHc had its coat protein replaced with hepatitis C vir黶 (HCV) core antigen gene. Anotherr TMV mutant TSBD was repli-case-defective. Coinfection of the two mutants could cause systemic infection in tobacco plants by in trans complementation of their functions. TSHc could effectively replicate and assemble to viral particles, which were a little longer than that of wild-type TMV. HCV core antigen was expressed in whole tobacco plants. A similar expression level of HCV core antigen was detected on serial passages, which suggested that this viral expression system be stable.