mlpack  blog
Deep Learning Module in mlpack - Week 9

Deep Learning Module in mlpack - Week 9

Kris Singh, 3 August 2017

Week Nine

Most of the week was spent in just fixing both the ssRBM and GAN PR. As the present situation stands ssRBM is achieving accuracy of 76% on the mnist dataset and around 79% on cifar10 dataset(500 images).I also made changes to the ssRBM PR so that now the input and output are now templated accepting any rather than just arma::mat. These accuracies are pretty good for cifar10 dataset but if still have to check for the whole dataset(10k) where the accuracy is ~68%

With the GAN PR I have modified the training algorithm now such that it only requires the use of only one Optimizer.optimize function call. Right now i am stuck with the GAN PR as after the training the outputs are essentially random and I was not able to find any mistakes with the training procedure. Also, there is no easy way to test GAN so i am confused as for how to test the implementation out other than just visualizing the outputs.

Next Week

The goal for the next week is to hopefully add some meaningful results from GAN PR. Also, I have started reading the stacked GAN paper it seems a bit of stretch if we would be able to implement it. I am hoping if GAN PR is complete by next week I can move on to stacked GAN.