中文摘要：

大数据是继云计算、物联网之后IT产业又一次颠覆性的技术革命,而深度学习推动了＂大数据＋深度模型＂时代的来临。本文是基于深度学习的稀疏自编码模型,把数理方程反问题中的新思想——同伦正则化的思想应用到该种模型中,将正则化参数的取值范围由无限区间（0,∞）改进为有限区间（0,1）,建立了一种新的深度学习的稀疏自编码模型。该模型与原模型相比较,实验设计简单,模型的正则化参数更加容易优化。

英文摘要：

Big data is a subversive revolution after cloud computing and Internet in IT industry.Deep learning drives the coming of the era of ＂ big data＋deep model＂.This article proposes a new sparse autoencoding model based on deep learning,the new idea,homotopy regularization,is applied to this model in inverse problem and the regularization parameter values range is improved from infinite interval（0,＋∞）to a limited range（0,1）.So this new sparse autoencoding model for deep learning is established.Compared with the original model,its advantage is that the regularization parameter of the model is easier to be optimized.