ArashAkbarinia / mbb

The teaching material for the deep learning course taught at the Mind Brain and Behaviour Master at JLU Giessen.

Introduction to Deep learning

The teaching material for the deep learning course taught at the Mind, Brain and Behaviour Master programme taught at JLU Giessen.

Instructor: Arash Akbarinia

I have compiled these materials into a Jupyter Book: https://deeplearning-jupyterbook.github.io/

1. Setting up a deep learning environment

The prerequisite to continue with the rest of the materials is to set up a deep learning environment. In this tutorial, we see how to do that using virtual environments.

2. Basic operations in artificial neural networks

Artificial neural networks consist of basic operations, such as convolution, pooling, and activation functions. In this session, we cover those operations.

• Assignment 1

3. Building a deep neural network project

In this session, we create a complete DNN project by constructing our own architecture and dataset. We train our network for a simple classification problem.

• Assignment 2

4. Optimisation and learning

In this session, we learn how a network acquires its knowledge and tunes its weights to perform a certain task by exploring different loss functions in toy examples of 2D points.

• Assignment 3

5. Vision

In this session, we learn about the task of semantic segmentation (i.e., having a label for each pixel). We train a network to perform this task and we look into transfer-learning.

6. Deep generative models

In this session, we learn about deep generative models that can learn the distribution of data to generate new samples. We will explore three major generative models:

• Generative Adversarial Networks (GAN)
• Variational Autoencoders (VAE)
• Diffusion Probabilistic Models (DPM)

7. Interpretation techniques

In this session, we learn about different interpretation techniques. How can we unravel the block box of deep neural networks? We will explore three techniques:

• Activation Map
• Kernel Lesioning
• Probing with Linear Classifiers

8. Python scripts for deep learning projects

In this session, we learn how to create a Python module that can be executed with a script on a GPU server. We explore the same code as last session (Probing with Linear Classifiers).

9. Other modalities

In this session, we move from vision to other modalities (audio and text) and learn how to use deep networks for these modalities in a simple classification problem. We also look at a multi-modal network:

• Audio
• Text
• Language – Vision

10. Reinforcement learning

In this session, we learn about the reinforcement learning paradigm in which an agent interacts with the environment and learns what the best policy is.