近年来,机器学习在计算机视觉中取得了许多突破性的研究进展.然而,已训练好的学习模型难以直接应用于相似但具有不同数据分布特征的其它学习任务中.域自适应技术通过抽取源域与目标域数据之间的公共特征,来实现把源域中学习到的知识迁...近年来,机器学习在计算机视觉中取得了许多突破性的研究进展.然而,已训练好的学习模型难以直接应用于相似但具有不同数据分布特征的其它学习任务中.域自适应技术通过抽取源域与目标域数据之间的公共特征,来实现把源域中学习到的知识迁移至目标域,从而避免针对目标域的训练数据收集和模型训练代价.但是,现有的视觉域自适应方法大都无法处理高阶的特征数据,一般都是通过简单的向量化操作将高阶张量特征转换成高维一阶向量特征.这不仅会破坏高阶特征数据内部的结构信息,而且还会增加算法的计算复杂度.为了解决上述问题,本文在保持原有张量特征结构不变的条件下,利用张量乘操作,将视觉域自适应问题抽象为求解源域和目标域的共同张量子空间以及源域和目标域特征在该共同张量子空间上投影的多变量优化问题.然后,利用张量奇异值分解和交替方向乘子法,提出一种基于张量奇异值分解的视觉域自适应方法(Visual domain Adaptation method based on TEnsor Singular value decomposition,VATES),以实现上述多变量优化问题的迭代求解.文中证明了正交张量子空间约束条件下源域与目标域表征误差最小化问题的可解性问题,并求得了相应的解析解.在公开数据集Office-Caltech-10、Office31、ImageNet-VOC2007上与17个基线模型进行对比实验.结果表明本文所提出的方法与经典的机器学习方法、非深度域自适应方法、深度域自适应方法以及张量域自适应方法相比,在无标签目标域上的图像分类精度分别提高了10.6%~43.9%、0.7%~31.1%、0.7%~24.8%以及5.7%~34.9.同时,算法的运行效率也提高了40.5%~74.3%,显著优于所对比的基线方法.实验分析也表明,VATES方法的目标域分类精度会随着所选用神经网络特征抽取能力的增强而逐渐提升.展开更多
Phytophthora root and stem rot of soybean caused by Phytophthora sojae(P.sojae)is a devastating disease that affects soybean[Glycine max(L.)Merr.]all over the world.S-phase kinase-associated protein 1(SKP1)proteins ar...Phytophthora root and stem rot of soybean caused by Phytophthora sojae(P.sojae)is a devastating disease that affects soybean[Glycine max(L.)Merr.]all over the world.S-phase kinase-associated protein 1(SKP1)proteins are key members of the SKP1/Cullin/F-box protein(SCF)ubiquitin ligase complex and play diverse roles in plant biology.However,the role of SKP1 in soybean against the phytopathogenic oomycete P.sojae remains unclear.In this study,a novel member of the soybean SKP1 gene family,GmSKP1 which was significantly induced by P.sojae,was reported.The expression of GmSKP1 was simultaneously induced by methyl jasmonate(MeJA),salicylic acid(SA)and ethylene(ET),which might suggest an important role for GmSKP1 of plant in responses to hormone treatments.Functional analysis using GmSKP1 overexpression lines showed that GmSKP1 enhanced resistance to P.sojae in transgenic soybean plants.Further analyses showed that GmSKP1 interacted with a homeodomain-leucine zipper protein transcription factor(GmHDL56)and a WRKY transcription factor(GmWRKY31),which could positively regulate responses to P.sojae in soybean.Importantly,several pathogenesis-related(PR)genes were constitutively activated,including GmPR1a,GmPR2,GmPR3,GmPR4,GmPR5a and GmPR10,in GmSKP1-OE soybean plants.Taken together,these results suggested that GmSKP1 enhanced resistance to P.sojae in soybean,possibly by activating the defense-related PR genes.展开更多
文摘近年来,机器学习在计算机视觉中取得了许多突破性的研究进展.然而,已训练好的学习模型难以直接应用于相似但具有不同数据分布特征的其它学习任务中.域自适应技术通过抽取源域与目标域数据之间的公共特征,来实现把源域中学习到的知识迁移至目标域,从而避免针对目标域的训练数据收集和模型训练代价.但是,现有的视觉域自适应方法大都无法处理高阶的特征数据,一般都是通过简单的向量化操作将高阶张量特征转换成高维一阶向量特征.这不仅会破坏高阶特征数据内部的结构信息,而且还会增加算法的计算复杂度.为了解决上述问题,本文在保持原有张量特征结构不变的条件下,利用张量乘操作,将视觉域自适应问题抽象为求解源域和目标域的共同张量子空间以及源域和目标域特征在该共同张量子空间上投影的多变量优化问题.然后,利用张量奇异值分解和交替方向乘子法,提出一种基于张量奇异值分解的视觉域自适应方法(Visual domain Adaptation method based on TEnsor Singular value decomposition,VATES),以实现上述多变量优化问题的迭代求解.文中证明了正交张量子空间约束条件下源域与目标域表征误差最小化问题的可解性问题,并求得了相应的解析解.在公开数据集Office-Caltech-10、Office31、ImageNet-VOC2007上与17个基线模型进行对比实验.结果表明本文所提出的方法与经典的机器学习方法、非深度域自适应方法、深度域自适应方法以及张量域自适应方法相比,在无标签目标域上的图像分类精度分别提高了10.6%~43.9%、0.7%~31.1%、0.7%~24.8%以及5.7%~34.9.同时,算法的运行效率也提高了40.5%~74.3%,显著优于所对比的基线方法.实验分析也表明,VATES方法的目标域分类精度会随着所选用神经网络特征抽取能力的增强而逐渐提升.
基金Supported by the NSFC Projects(31971972)the Natural Science Foundation of Heilongjiang Province(ZD2019C001)the Outstanding Talents and Innovative Team of Agricultural Scientific Research。
文摘Phytophthora root and stem rot of soybean caused by Phytophthora sojae(P.sojae)is a devastating disease that affects soybean[Glycine max(L.)Merr.]all over the world.S-phase kinase-associated protein 1(SKP1)proteins are key members of the SKP1/Cullin/F-box protein(SCF)ubiquitin ligase complex and play diverse roles in plant biology.However,the role of SKP1 in soybean against the phytopathogenic oomycete P.sojae remains unclear.In this study,a novel member of the soybean SKP1 gene family,GmSKP1 which was significantly induced by P.sojae,was reported.The expression of GmSKP1 was simultaneously induced by methyl jasmonate(MeJA),salicylic acid(SA)and ethylene(ET),which might suggest an important role for GmSKP1 of plant in responses to hormone treatments.Functional analysis using GmSKP1 overexpression lines showed that GmSKP1 enhanced resistance to P.sojae in transgenic soybean plants.Further analyses showed that GmSKP1 interacted with a homeodomain-leucine zipper protein transcription factor(GmHDL56)and a WRKY transcription factor(GmWRKY31),which could positively regulate responses to P.sojae in soybean.Importantly,several pathogenesis-related(PR)genes were constitutively activated,including GmPR1a,GmPR2,GmPR3,GmPR4,GmPR5a and GmPR10,in GmSKP1-OE soybean plants.Taken together,these results suggested that GmSKP1 enhanced resistance to P.sojae in soybean,possibly by activating the defense-related PR genes.
