作 者: ();
机构地区: 华南理工大学工商管理学院
出 处: 《信息技术》 2020年第4期5-9,共5页
摘 要: 自编码器是深度学习中广泛使用的一种网络,许多深度网络都以其为基础构建特定的网络结构。对于传统自编码器及其变体,不同层单元之间的参数被限定为普通实数,这在某种程度限制了自编码器的能力,为了克服这种问题,文中以模糊理论为基础,通过将自编码器的参数设置为模糊数,构建了模糊自编码器模型,并引入了相应的学习算法。基于MINIST手写数字数据集的实验表明,模糊自编码器在表示能力和鲁棒性方面都要优于普通自编码器。 Autoencoder is a type of network which is widely used in deep learning,and many deep networks build specific network structures based on them.For the traditional Autoencoder and its variants,the parameters between different layer units are limited to ordinary real numbers,which limits the ability of the Autoencoder to some extent.In order to overcome this problem,based on fuzzy theory,the parameters of the autoencoder are set to fuzzy numbers,a fuzzy autoencoder model is constructed.Meanwhile the learning algorithm corresponding to fuzzy autoencoder is introduced.Experiments based on MINIST handwritten digital data sets show that fuzzy autoencoders are superior to regular autoencoders in terms of representation ability and robustness.
领 域: []