This is my work of using Gumbel Softmax to modify the Gaussian Mixture Variational Autoencoder. You can find the technical report here. The work is supervised by Prof. Zhiqiang Ge(scholar) and Dr. Zhichao Chen.
The repository also contains the unfinished work of DPMM, in which I tried to infer the number of components by using a Direchlet Process. It sounds like an elegant idea, but the performance of model using DPMM is disappointing. I may take a look at this later.