fast.ai course for Google Colab
Fast.ai is an online platform on which one can learn deep learning by implementing the PyTorch library. It consists of 14 lectures — divided into 2 parts, with 7 lectures each. Unlike the reputed Deep Learning course by Dr. Andrew Ng on Coursera, the fast.ai course implements a top-down approach — which means instead of focusing on the theory behind each algorithm, fast.ai gives a hands-on experience of how to build and use DL models. While using these models, you will learn necessary topics when needed.
This course, however, requires students to make an account on websites such as Crestle or Paperspace to run the code provided in the course as it requires a GPU-enabled system. Since everyone does not have the resources to make such an account (mostly due to these services requiring a credit card), or build an equally powerful system at home, the audience of this course is limited.
Keeping these factors in mind, I went ahead and modified the code in the python notebooks to make it compatible with Google Colab.
What is Google Colab?
Google Colab or Colaboratory, is a free Google cloud service for AI researchers and developers. They can run their code from the cloud and store it on their Google drive. Google also added Tesla K80 GPUs to Colab with which we can run our projects even faster.
The Colab-compatible fast.ai code can be found here.
The code is pretty self-explanatory but I’ve added a small explanation on what each line does. Feel free to check it out and point out any error in the code. Cheers!
Article by Sriram Vijendran, Researcher in Minsky Lab at Next Tech.
