本文针对小样本图像分类问题,提出一种基于样本对的元学习(Pairwise-based Meta Learning,PML)方法.利用传递迁移学习对预训练好的Resnet50模型进行微调,得到一个更适应小样本任务的特征编码器,将该特征编码器作为元学习模型的初始特征...本文针对小样本图像分类问题,提出一种基于样本对的元学习(Pairwise-based Meta Learning,PML)方法.利用传递迁移学习对预训练好的Resnet50模型进行微调,得到一个更适应小样本任务的特征编码器,将该特征编码器作为元学习模型的初始特征编码器来训练模型,进一步增强了元学习模型的泛化能力;同时,本文还基于支持集与查询集样本之间的相似性提出元损失函数(Meta Loss,ML),其考虑了特征空间中查询集所有样本的相互关系,以此来缩小正样本类内距离,增加正负样本类间距离,从而提高分类精度.实验结果表明,本文的方法在1-shot、5-shot任务上分别达到了77.65%、89.65%的分类精度,较最新的元学习方法Meta-baseline分别提高7.38%、5.65%.展开更多
1. There are fruitful results in geometrical function theory of one complex variable. But there exist a lot of counter examples to show that the corresponding results in several complex variables are not true. For the...1. There are fruitful results in geometrical function theory of one complex variable. But there exist a lot of counter examples to show that the corresponding results in several complex variables are not true. For the classical distortion theorem in one complex variable, H. Cartan conjectured that it is valid for the biholomorphic mappings on the unit ball B^n in C^n (n≥2). Unfortunately, this conjecture is not true(see [2]). In this report, we will consider the distortion theorem of biholomorphic mappings on transitive do-展开更多
文摘本文针对小样本图像分类问题,提出一种基于样本对的元学习(Pairwise-based Meta Learning,PML)方法.利用传递迁移学习对预训练好的Resnet50模型进行微调,得到一个更适应小样本任务的特征编码器,将该特征编码器作为元学习模型的初始特征编码器来训练模型,进一步增强了元学习模型的泛化能力;同时,本文还基于支持集与查询集样本之间的相似性提出元损失函数(Meta Loss,ML),其考虑了特征空间中查询集所有样本的相互关系,以此来缩小正样本类内距离,增加正负样本类间距离,从而提高分类精度.实验结果表明,本文的方法在1-shot、5-shot任务上分别达到了77.65%、89.65%的分类精度,较最新的元学习方法Meta-baseline分别提高7.38%、5.65%.
文摘1. There are fruitful results in geometrical function theory of one complex variable. But there exist a lot of counter examples to show that the corresponding results in several complex variables are not true. For the classical distortion theorem in one complex variable, H. Cartan conjectured that it is valid for the biholomorphic mappings on the unit ball B^n in C^n (n≥2). Unfortunately, this conjecture is not true(see [2]). In this report, we will consider the distortion theorem of biholomorphic mappings on transitive do-
