A Deep Neural Network Architecture Using Dimensionality Reduction with Sparse Matrices
A Deep Neural Network Architecture Using Dimensionality Reduction with Sparse Matrices
复制标题
使用稀疏矩阵降维的深度神经网络架构
DOI：
10.1007/978-3-319-46681-1_48
复制
发表时间：
2016
期刊：
影响因子：
--
通讯作者：
Z. Xiongxin
中科院分区：
文献类型：
--
作者：
W. Matsumoto;Manabu Hagiwara;P. T. Boufounos;K. Fukushima;T. Mariyama;Z. Xiongxin
我们提出了一种新的深度神经网络架构，其动机是稀疏随机矩阵理论，该理论通过稀疏矩阵而不是传统的堆叠式自编码器使用低复杂度嵌入。我们认为自动编码器是一种信息保持的降维方法，类似于压缩感知中的随机投影。因此，利用最近的理论稀疏矩阵降维，我们实验证明，分类性能不会恶化，如果自动编码器被替换为一个计算效率高的稀疏降维矩阵。
We present a new deep neural network architecture, motivated by sparse random matrix theory that uses a low-complexity embedding through a sparse matrix instead of a conventional stacked autoencoder. We regard autoencoders as an information-preserving dimensionality reduction method, similar to random projections in compressed sensing. Thus, exploiting recent theory on sparse matrices for dimensionality reduction, we demonstrate experimentally that classification performance does not deteriorate if the autoencoder is replaced with a computationally-efficient sparse dimensionality reduction matrix.