文件列表:
通过流形平展和重构进行表示学习【英文版】.pdf |
下载文档 |
资源简介
>
英文标题:Representation Learning via Manifold Flattening and Reconstruction中文摘要:本研究提出了一种算法,可从流形数据样本中构建一对神经网络,以实现流形的线性化和重建,这些网络被称为 Flattening Networks(FlatNet),能够在理论上解释,可在大规模上可行,且在测试数据上具备很好的泛化能力,在高维流形数据和二维图像数据上进行了实证结果和比较,代码已公开发布。英文摘要:This work proposes an algorithm for explicitly constructing a pair of neuralnetworks that linearize and reconstruct an embedded submanifold, from finitesamples of this manifold. Our such-generated neural networks, called flatteningnetworks (FlatNet), are theoretically in
加载中...
本文档仅能预览20页