Syllabus

• Week 1

  Introduction to deep learning

  • Machine learning versus deep learning

    A recap of what we mean by machine learning, and an introduction to the core concepts of deep learning, including convolutions and representation learning

  • Deep learning libraries and PyTorch

    An introduction to some of the software libraries used for deep learning, in particular the one we will be using on this course, PyTorch.

  • Summary and review

    A summary of week 1 of the course, Introduction to Deep learning, with a review and quiz.

• Week 2

  Convolutional neural networks

  • Network components

    A closer look at the building blocks of deep learning networks, including convolution layers, max pooling layers, and activation functions.

  • Tensors and dimensionality

    Data is passed between the layers of a CNN in the form of tensors. What are tensors and how are they made and manipulated in PyTorch?

  • A simple CNN

    An overview of how to build a Convolutional neural network, plus a step by step implementation of a simple CNN using PyTorch.

  • Summary and review

    A review of Week 2, our tour of the building blocks of CNNs, including a quiz and a look ahead to next week.

• Week 3

  Training CNNs

  • The training loop

    A look at the steps performed during a typical training loop, including loss functions, forward and backward passes, optimisation, with an example in PyTorch.

  • Hyperparameters

    Network performance can be improved via the adjustment of hyperparameters. We discuss some important hyperparameters including learning rate, and network depth and shape.

  • Datasets and dataloaders

    Deep learning requires a lot of data. Data augmentation can artificially increase the size of datasets, while dataloaders can help deep learning networks access that data efficiently.

  • Summary and review

    A review of week 3, training convolutional neural networks, including a quiz

• Week 4

  Classification and regression

  • Encoder architectures

    A look at the general structure or architecture of deep classifiers, plus an overview of some of the most commonly used encoder architectures.

  • Making an image classifier in PyTorch

    Practical exercies showing how to make custom datasets in PyTorch, and train and test an image classifier.

  • Regression using deep learning

    A look at how deep learning can be used for regression problems, including a practical example

  • Summary and review

    A look back at week 4 of the course, on classification and regression using deep learning.

• Week 5

  Spatial approaches and more

  • Encoder-decoder architectures

    Concepts, architectures and loss functions associated with encoder-decoder networks.

  • More advanced topics

    A brief look at some more advanced topics and applications of applications of deep learning, including heatmap regression, bounding box methods, and multi-objective learning.

  • Summary and review

    A recap of our overview of spatial deep learning approaches, and of the course in general.

Who will you learn with?