search

ericjang / gumbel-softmax

categorical variational autoencoder using the Gumbel-Softmax estimator
MIT License
425 stars 101 forks source link

About Unsupervised Clustering Under VAE？ #6

Closed LynnHo closed 4 years ago

LynnHo commented 6 years ago

@ericjang Thanks for the good code! I wonder whether we could use gumbel-softmax for clustering under VAE framework (e.g. for Mnist, 392->..->10->..->392, and "10" represents the relaxed categorical distribution). I have tried, but got bad results. Do you have any idea?

akihironitta commented 4 years ago

I guess that feature representations learnt by squashing input, e.g. images, into only 10 dimensional categorical variable using encoder-decoder framework will be very sensitive to initial weights of NN because no label information is available in unsupervised learning tasks like clustering.