/thinking-deep-learning-tutorial

Repo for material and instructions for the Thinking with Deep Learning tutorial.

Primary LanguageJupyter NotebookMIT LicenseMIT

Thinking Deep Learning Tutorial

Hello! Welcome to the thinking with deep learning tutorial. We will be largely running this tutorial through Google Colab, but there is also a Jupyter notebook you could use locally.

The purpose of this tutorial is to introduce deep learning, broadly, to practitioners without a necessarily strong background in machine learning. It is especially useful to those working with text, images, graph/network or tabular data, from a social scientific or natural science context, but will be generally handy when performing any data analysis.

We will start with the two major deep learning frameworks, Keras and PyTorch, before moving into different data specific examples, with a few illustrations using social data.

Link to Tutorial Slides

You can follow along the slides over here: https://drive.google.com/file/d/1SYAh6pyV3TVLnTT9_-Iw78I7N_D6RN7W/view?usp=drive_link

Tutorial setup and Instructions

The tutorial will be smoothest on Colab, as you will not have to bother with individual package installations. On local, you are expected to have Keras and PyTorch. There are domain specific packages that you should be easily able to grab from PyPi, but as it is a large tutorial, you might not be running everything.

This is the link to the Colab tutorial: https://colab.research.google.com/drive/1vEB2u41LxQ8eC1J3o_c6J_T_ua5YSxv2?usp=sharing

Once you open it, the first thing you would want to do is to create a copy of it in drive (under the file tab), so that you can edit as you would like. If you have colab pro, we recommend using a GPU while running the notebook.

NOTE: Some large-ish model downloads!

At different parts of the tutorial, we will be downloading pre-trained text, image and audio models - this might take some time, so if you would like, you can set up the colab notebook before the tutorial starts, search "download", and get your models ready.

Additional Material and Resources

Time constraints mean that we are rarely able to get to showcasing the full capabilities of these approaches - luckily for you, the material is an abrdiged version of a 10 week course, and an upcoming textbook titled "Thinking with Deep Learning".

In this repository, you will find the homework exercises for the Spring 2022 version of the course, with added material in there for you to explore. Each week covers a different modality or domain of deep learning applied to social scientific or natural science datasets.

This is a link to the the github repo where you can also find data: https://github.com/Thinking-with-Deep-Learning-Spring-2022