Assignments for the Spring 2019 Deep Learning course at University of Amsterdam

1. Derived analytical gradients for the FC and Batch Normalization layers
2. Numpy implementation of the FC and activation functions with custom backward methods
3. Pytorch implementation of the BatchNorm module with custom backward method
4. Trained MLP and simple CNN on the Cifar10 dataset

Find report with derivations here

Find code here here

1. Derived analytical gradients for the RNN
2. Pytorch implementation of the vanilla RNN and LSTM trained to memorize palindromes with various length

3. Trained LSTM as a generative model to predict the next character in the text

Find report with derivations here

Find code here here

1. Pytorch implementation of VAE

ELBO for VAE with 20-dim latent space

Images sampled from Decoder at the begining, halfway through, and at the end of training

Output of VAE’s decoder in 2-dimensional latent space

2. Pytorch implementation of GAN

Loss of Generator and Discriminator networks

Images sampled from Generator at the begining, halfway through, and at the end of training

Interpolating between two images in the latent space

3. Pytorch implementation of RealNVP

Training and validation performance in bits per dimension

Images sampled from RealNVP after the 1-st epoch, at the middle point of training, and at the end of training

Find report here

Find code here here