本文为加拿大多伦多大学（作者：Tijmen Tieleman）的博士论文，共120页。

图像识别又称计算机视觉，是神经网络最突出的应用之一。本文提出的图像识别方法是基于逆向过程：生成图像。对于我们今天的计算机系统来说，生成图像比识别它们要容易。这项工作利用了生成图像的能力，以便识别其他图像。本文的一部分介绍了这种“综合分析”思想在一个复杂自动编码器中的具体实现。图像生成系统的一半（即系统的结构）是硬编码的；另一半（结构中的内容）是学习的。在学习该图像生成系统的同时，附带的图像识别系统正在学习从图像中提取描述信息。通过共同学习，这两个部分对所提供的数据有了很好的理解。本文的第二部分是一个无向生成模型的训练算法，它利用训练和马尔可夫链之间强大的交互作用，马尔可夫链的任务是从模型中产生样本。该算法在处理图像数据时表现良好，但同样适用于其他类型数据的无向生成模型。

Image recognition, also known as computervision, is one of the most prominent applications of neural networks. The imagerecognition methods presented in this thesis are based on the reverse process:generating images. Generating images is easier than recognizing them, for thecomputer systems that we have today. This work leverages the ability togenerate images, for the purpose of recognizing other images. One part of thisthesis introduces a thorough implementation of this “analysis by synthesis”idea in a sophisticated autoencoder. Half of the image generation system(namely the structure of the system) is hard-coded; the other half (the contentinside that structure) is learned. At the same time as this image generationsystem is being learned, an accompanying image recognition system is learningto extract descriptions from images. Learning together, these two componentsdevelop an excellent understanding of the provided data. The second part of thethesis is an algorithm for training undirected generative models, by making useof a powerful interaction between training and a Markov Chain whose task is toproduce samples from the model. This algorithm is shown to work well on imagedata, but is equally applicable to undirected generative models of other typesof data.

1. 文献回顾：深度神经网络

2. 具有域特定解码器的自动编码器

3. 训练无向生成模型

4. 附录：详细推导过程

更多精彩文章请关注公众号：