Taking the Course: Practical Deep Learning for Coders

This blog post describes the content, value, and experience of taking Lesson 1 of the Practical Deep Learning for Coders course. The post also includes the lesson video.

Taking the Course

Last week I started taking an online course called Practical Deep Learning for Coders. My motivation was based on personal interest, but if you are a practicing data scientist, this course would be valuable to your professional development. It provides a foundational understanding of deep learning which will help you assess whether you’d like to incorporate deep learning into your work.

Uses for Deep Learning

Deep learning is an exciting, rapidly-developing field that has a wide range of applications. There are many uses for deep learning. It can help scientific researchers gain insights into fields like botany, drug discovery, and agriculture. Deep learning is also being used for fraud detection, speech recognition, and to build “emotion aware” apps. The deep learning course that I took was developed by Rachel Thomas and Jeremy Howard of fast.ai. It is free and it has a robust online community centered around it.

Lesson 1 Synopsis

From the very beginning, it is clear that this course is not a typical educational experience. There is an emphasis on making the material accessible to a broad audience and getting to the fun stuff as soon as possible.

Lesson 1 starts from scratch and ends with you submitting an image recognition model to a Kaggle competition. The lecture and the homework exercises walk you through foundational tasks. Along the way, there is enough information and homework practice for those, like me, that can’t do such things in their sleep. Yet, everything moves along at a good pace. Pretty soon you are thinking about data structuring in a whole new way and you are getting friendly with NumPy, Keras, and Matplotlib. Then things get really fun as you start using and then fine tuning a model called VGG16 to distinguish between images of cats and images dogs. Then you have your work evaluated by entering a Kaggle contest.

It was very rewarding to be actually using deep learning after just one lesson. I did not need to refresh my linear algebra skills before I was using a working, complete deep learning model on real photographs. Though, if and when I do want to refresh my linear algebra skills, fast.ai has a free, online textbook and video series for that.

Summary

I deeply appreciate that the fast.ai founders created this educational resource for a broad audience that includes myself and data scientists. I am also grateful that fast.ai and USF Data Institute created the Diversity Scholarship which introduced me to Dawit Haile. Dawit is doing the in-person version of the deep learning course at the Data Institute. Look for blog posts summarizing our experiences from both me and Dawit once we have each completed all the lessons.

Domino editorial note: classroom image provided courtesy of USF