Exercise 1: Introduction to Deep Learning

Overview

In this exercise we will explore the basic building block of deep learning: the perceptron, and how to stack multiple perceptrons together into layers to build a neural network. We will also introduce convolutional neural networks (CNNs) for image classification.

In particular, we will:

• Introduce PyTorch, a popular framework for deep learning.
• Implement a perceptron and a 2-layer perceptron to compute the XOR function using PyTorch.
• Implement and train a simple neural network (a multi-layer perceptron) to classify points in a 2D plane using PyTorch.
• Implement and train a simple deep convolutional neural network to classify hand-written digits from the MNIST dataset using PyTorch.
• Discuss important topics in ML/DL, such as data splitting, under/overfitting and model generalization.

Setup

Please run the setup script to create the environment for this exercise.

source setup.sh

You can now open the exercise.ipynb file in VSCode. Please make sure that the Python and Jupyter VSCode extensions are installed before proceeding with the exercise. When you are ready, simply follow the instructions in the notebook from the beginning.