Deep Learning 1 Course - Practicals

This repository contains the code part of the three assignments of the Deep Learning 1 course, Fall 2021 edition. I am omitting my University name for searchability reasons. My MSc university can be found on my LinkedIn or CV.

Assignments

More details for each assignment can be found in the assignment pdfs. For a brief overview, refer to the following:

1. Assignment 1: MLPs and Backpropagation. The following is implemented:
   • Differentiable Cross Entropy in NumPy
   • Differentiable Softmax in NumPy
   • Differentiable ReLU in NumPy
   • Differentiable Linear Layer in NumPy
   • A Multi-Layer Perceptron (MLP) in NumPy
   • An MLP in PyTorch
   • Training and Evaluation of both MLPs on CIFAR10
2. Assignment 2: CNNS, RNNs, and GNNs. The following is implemented:
   • Part 1: CNNs
     • Building blocks of a convolutional neural network in NumPy
       • Zero padding in NumPy
       • Differentiable convolution in NumPy
       • Differentiable Max Pooling in NumPy
     • Training and evaluation of a number of torchvision models (ResNet-{18,34}, VGG-11, DenseNet-121)
   • Part 2: RNNs
     • LSTM in PyTorch, using only nn.Parameter and non-linear activation functions
     • Training and evaluation of generative LSTM Language Model on books.
   • Part 3: GNNs
     • Implementation of Graph Convolutional Neural Networks trained and evaluated on molecule data.
3. Assignment 3: Variational Autoencoders
   • Implementation of a Convolutional Variational Autoencoder in PyTorch
   • Training and Evaluation on FashionMNIST generation.