What is it about?

We devised a novel generative network using self-organizing map. Our network can be used as a general generative model like VAE and GAN. The advantages of our network compared to VAE and GAN are as follows. Our network can make more clear images than VAE because there is no sampling in training. Our network can be easily trained without mode collapsing problems unlike GAN, so application is very easy. The one disadvantage is that the training time of self-organizing map is considerably increased as the dimension is increased.

Featured Image

Why is it important?

Nowadays, generative models have a lot of application domains, but existing networks have some drawbacks. Our method is a new approach for generative model, so it provides a new branch to this researh area. Moreover, our method can be expanded to many application such as voice generation and natural language processing.

Read the Original

This page is a summary of: SOGN: A Novel Generative Model using Self-Organizing Map, Electronics Letters, March 2019, the Institution of Engineering and Technology (the IET), DOI: 10.1049/el.2019.0202.
You can read the full text:

Read

Resources

Contributors

The following have contributed to this page