Introspective Neural Networks for Generative Modeling
Introspective Neural Networks for Generative Modeling
复制标题
DOI：
10.1109/iccv.2017.302
复制
发表时间：
2017-10
期刊：
影响因子：
--
通讯作者：
Justin Lazarow;Long Jin;Z. Tu
中科院分区：
文献类型：
--
作者：
Justin Lazarow;Long Jin;Z. Tu
我们通过开发一个由渐进学习的深度卷积神经网络构建的生成模型来研究无监督学习。由此产生的生成器是一个额外的训练器，在某种意义上能够“内省”-能够自我评估其生成的样本和给定的训练数据之间的差异。通过重复的判别学习，现代判别分类器的理想属性直接由生成器继承。具体来说，我们的模型使用分类合成算法学习一系列CNN分类器。在实验中，我们观察到令人鼓舞的结果，包括纹理建模，艺术风格转移，人脸建模和无监督特征学习的应用。
We study unsupervised learning by developing a generative model built from progressively learned deep convolutional neural networks. The resulting generator is additionally a discriminator, capable of "introspection" in a sense — being able to self-evaluate the difference between its generated samples and the given training data. Through repeated discriminative learning, desirable properties of modern discriminative classifiers are directly inherited by the generator. Specifically, our model learns a sequence of CNN classifiers using a synthesis-by-classification algorithm. In the experiments, we observe encouraging results on a number of applications including texture modeling, artistic style transferring, face modeling, and unsupervised feature learning.