Deep learning is a vast field of machine learning. Despite all the hype and promise it receives, it can be a daunting task for a beginner who is interested in venturing into deep learning. Before I got a solid footing, I struggled for quite sometime to understand what deep learning was.
My first mistake was to pick a machine learning book. Why do I say this was a mistake? Well, deep learning is a subset of machine learning, hence they share some similaries but the core difference is how machine learning and deep learning algorithms work. If you have ever read a machine learing book or blog post, a common dominator is feature engineering.
For this sake of keeping this post short, I ask you to dig into the web to find more about feature engineering. Remember when I mentioned it was a mistake to pick a machine learning book; turns out deep learning really does not care much about feature engineering. Everything is in the data, hence the very essence of deep learning. I took my machine learning knowledge and tried to use it to on a deep learning problem. I did not get far.
After weeks of many trials and failures, I stumbles upon a very good book that opened my eyes. Before then, I did not have a good understanding of what deep learning was. In my head, it was all machine learning. Well, true but not quite.
Anyway I would like to present to you a list of resources that helped shape my understanding of deep learning and its place in AI.
1 . Deep learning in Python by François Chollet
This is by far the most valuable resource. Not only was it written by the author of Keras, but it intuitively builds your knowledge regardsless of previous background. It very little mathematical notions which can be a turn off. I highly recommend it.
2 . Deeplizard youtube channel
Similar to above, the presenter assumes no previous knowledge. Explainations are very easy to follow and the videos are quite short and condensed with alot of valuable insights.
3 . Tensorflow documentation
As obvious as it may seem, I’m one of those people who rarely read documentations. It was during my search for answers that I found out the documentation was very informative and detailed. It drastically reduced the time I would search on Stackoverflow looking for answers.
4 . Source code
Many a times I have found myself reading the source code to understand the mechanics and how things fit in together. This is especially true when I implemented a custom loss function. Although it did not work as I expected, it gave me an understanding of the internals.
5 . Google AI blog
I believe Google as company has made huge strides when it comes to AI and its products are a testament to this statement. Hence, its only wise to read what trendsetters are doing and how they are doing it. Though most of the posts are rather technical, the take away is the abstract idea that the blogs present in solving a variety of AI related problems. I recommend this if you want to know where the industry is heading. Most of them have links to research papers whose read may of be interest especially for research purposes or trying out new architectures.