“The illiterate of the 21st century will not be those who cannot read and write, but those who cannot learn, unlearn, and relearn.” (Alvin Toffler)

This is a quote I found on a blog named "Through Tinted Lenses" written by a fellow deep learning student named Hasish Vadlamani. It not only explains what process is required to stay competitive in the 21 century, but also the deep learning process itself. As I embark on this journey I must learn, unlearn and relearn everything -- just like a deep learning model in the field of AI and machine learning.

I have been working in the vfx and animated film industry for many years as a cg artist, pipeline developer and a cg superviror. I have always been interested in technology and state-of-the-art advancements in tech since I was young. As AI and deep learning started impacting my life in more and more ways, I became intrigued and suddenly I wanted to learn how to leverage the power of AI and deep learning myself as I wanted to learn, unlearn and relearn what I could do with these powerful tools.

Off The Deep End

Learning Deep Learning is hard for many obvious and not so obvious reasons.

Some of the obvious reasons they are state-of-the-art -- meaning the advances in these areas happening at a break-neck speed. Other obvoius reasons is that they use high level math and complex algorithms at their core. Lastly, the general public believes that AI and machine learning are relegated to research labs and PHD practictioners who understand all the theories, math and algorithms on a fundemental level. All of these make the thought of learning AI and deep learning a daunting task to undertake.

As I started my journey into AI and deep learning, I found two not so obvious reasons why learning about AI and deep learning are hard. First, I found that institutions of higher learning - like all types of businesses acted as gatekeepers guarding their knowledge and instruction behind ivory towers of higher education and behind pay walls of online classes. Second, most other free websites or sources on the subject approached the topic in the traditional bottom up approach where you learn the low level mathmatics, and heavy theory first then gradually build up to leanring anything practical that could be applied to real world cases.

I could not go down these routes as a self learner. I did not have the funds time, or energy to go these traditional rountes of educating myself. I kept on starting and stopping on the subject and felt like I was moving nowhere.

Finding Fastai

“I hear and I forget. I see and I remember. I do and I understand.” (Chinese proverb)

It was a conversation with a friend and co-worker of mine that originally lead me to the Fastai website. My friend, like myself had started to learn about the topic, but a year or two ealier than me, but unlike me he had successfully used a state-of-the-art deep learning model on a recent project of his. He stopped me in my tracks while talking about my frustrations in learning about the topic and suggested I look at the Fastai website.

I took his advice and was immediately intrigued. The overall philosopy of Fastai was simple - "Making Neural Nets Cool Again".

I took his advice and was immediately intrigued. I binged watched the materials of the first course right away and felt like I was actually learning something practical that I could begin to put into practice right away. It was a big departure to what I had experienced before on my AI and deep learning journey and I was hooked. Like the proverb, I was understanding because I was doing.

Deep Learning Through Play

The philosphy of Fastai is simple. Democritize the tools and learning approach to deep learning so everyone so they might apply the principles easily to their own fields and domains.

The approach used directly challenges the usual way of learning by following a top-down approach to understanding deep learning. In comparison to the traditional bottom up approach used in academia. A top-down approach encourages you to jump right in and to implement models on your own very quickly. Students in the class learn to apply the theoritical concepts immediately using concrete state-of-the-art examples rather than bog them down in mathematical proofs. In a rapidly evolving field such as this one, being able to learn, experiment and prototype quickly is invaluable.

Just like you would not teach a child to play baseball by having them study math and physics for many years before picking up playing with a baseball, so to does Fastai hand you the ball first to play with before it leads you down into the more academic depths of the subject if you so wish.

There are very few prerequisists to start the course other than a high school math education and intermediate coding skills in Python, of which most can be picked up along the way. This, by no means a beginners only course. In fact, it is in depth enough that even long time practitioners in the field can learn and discover new things along the way. The course walks through new state-of-the art approaches, and techniques, tricks of the trade, and smart implementations that enable newly capable students to be able to compete in benchmark competitions such as Kaggle along with the biggest corporations and research groups who have much bigger resources and computer power.

Besides the online course, there is a corresponding book written by the founders of Fastai Jeremy Howard and Sylvian Gugger that follows the same philospy as the website.

The course is by no means lite version of the subject. It dives deep in certain topics that are required to become a proficient deep learning practitioner. Luckily, between Jeremy's informative lectures, the information and resources provided on the website, and the community of like-minded individuals makes the process learning about AI and deep learning attainable to almost anyone who has the courage to take the first steps. That is why I also recommend it to anyone who wants to learn more about AI and deep learning.

Why This Blog Series?

As I stated early in this article, my motivation comes down to wanting to learn more about AI and deep learning. While blogging may come across as a self-promoting endevor, I hope that by blogging about the subject, it will force me to become more proficient about the topics I write about. If anything it will give me and others another resource to help on the journey.

This approach of blogging to learn is not something I made up. I am taking the advice of Rachel Thomas(co-founder and co-author of fast.ai) who strongly encourages anyone who is starting a new learning path to put out blog posts about it in order to maximize your learning. The more I thought about her suggestion the more it made sense to start.

Besides the learning advantages, I hope to connect with other learners and or individuals who are also interested in the topics of AI and deep learning and I hope that we can learn from eachother along the way.

Learning about AI and deep learning can be daughting and overwhelming, but thanks to Jeremy Howard and Rachel Thomas the creators and authors of Fastai, that journey and adventure seems much more achievable that ever before.