Recently, I am reading some papers relating to unsupervised representation and convolutional network. The aim to write the blog is just to batter understand it. If there are misunderstandings or wrong ideas, please don’t hesitate to contact with me, which makes me full of grateful.
Why the authors want to write this paper?
In recent years, the convolutional networks(CNNs), a attracting method which owns good performance in image representation, has a wide range of applications in supervised learning which has seen huge adoption in computer vision. Comparatively, less attention has been payed to unsupervised learning with CNNs.
- What is the staring point or breakthrough point from which the authors deal with the above problem?
-
At first, the architecture called Deep Convolutional Generative Adversarial Networks (DCGANs) mainly based on attracting Generative Adversarial Networks (GANs) which provides effective alternative to maximum likelihood techniques instead of solving intractable probabilistic computations in which the maximum likelihood result during the training. However, GANs suffer from the difficult that the training is unstable, which lead Generators to produce the non-ideal outputs. - The architecture