Deep Learning Lab Assignments

Description:

This repository contains code for lab assignments of Deep Learning course offered in MSc. in Artificial Intelligence at the University of Amsterdam.

Assignemnt 1: MLP backprop, NumPy implementation, Pytorch MLP, Batch Normalization, Pytorch CNN

In this lab, we learn to implement and train basic neural architectures like Multi-Layer Perceptron (MLPs) and Convolutional Neural Networks (CNN) for image classification tasks. We also implement custom operations in PyTorch; a batch-normalization layer.

Problems and solution

• Main problem statement, Python Code and Report

Assignment 2: Vanilla RNN, LSTM, Modified LSTM Cell, Recurrent Nets as Generative Model

In this lab, we familiarize with Vanilla recurrent neural networks (RNNs) and LSTM on a simple toy problem of Palindrome Numbers. We study more advanced LSTM cell to emphasize the wide variety in LSTM cells. Finally, we use LSTM for learning and generating text.

Problems and solution

• Main problem statement, Python Code and Report

Assignment 3: Variational Auto Encoders and Generative Advesarial Networks

In this lab, we study and implement Deep Generative Models, specifically Variational Auto Encoders (VAE) and Generative Advesarial Networks (GAN).

Problems and solution

• Main problem statement, Python Code and Report

Results:

Training VAE


Manifold learned by the VAE


Interpolation results from GAN

Poster Presentations

We presented Attention is all you Need paper by Vaswani et al. This was done in collaboration with Vikrant Yadav and Tarun Krishna.

The poster can be found here.