The surprising fact from the very first page is the whole “I will just make an AI”, it isn’t as easy as it sounds. Because we often think that the machine will learn it all, and we (the programmer) just need to sit down and watch it learn. But the word machine learning is to some extinct misleading. The machine learns when you have given the machine a practical problem, which is the datasets. And algorithmically building some statistical model based on that dataset. And the statistical model will from that data begin to predict data from the model. And the statistical model is somehow what we call ‘learning’.
Now to make sense of it all, when humans learn. We do it in the four core learning styles. Which includes visual, auditory, reading, writing and kinesthetic.
The machine also learns in different ways, which includes. Supervised learning, Unsupervised learning, Semi-supervised learning, reinforcement learning.
Let’s take a look at supervised learning. To not make this boring. I like to think of machine learning as a person/persons (more on that later).
How does a machine learn?
But in supervised learning. If I gave some analogy to make sense of it all. You are supervised to learn from someone more experienced than you. That could be you (the student) and a teacher.
So the teacher would say: What’s 9+10?
Me: its 3.
Teacher: Not even close it’s 19.
Me : it’s 16.
Teacher : Wrong its still 19.
Me : Its 18.
Teacher : No, its 19.
Me : it’s 19.
Teacher : Good job, here is your grade. (A+)
But supervised learning is the most common type of learning in machine learning. This is how a machine learns to label some emails as spam or not.
Now, unsupervised learning speaks to itself. Without someone there telling you it is wrong or right. So you become a self-learner.
supervised kid: My teacher gave me so much to read and so many tests!
Unsupervised kid: Me too. But at least they told you the answers! ****
You can by now make sense of the name. But Semi-supervised learning refers to a machine that occasionally gets supervised by the teacher. And the teacher doesn’t know much. The only thing that teacher know is (if we take the example from before) what a banana looks like, but the teacher doesn’t know what an apple is. So the machine can group the food into two things banana, and non-banana.
Reinforcement means according to the dictionary “The process of encouraging or establishing a belief or pattern of behaviour. “
So to make sense of this type of learning. Imagine a student with no prior knowledge or nor a teacher. The only thing it knows is the environment it lives in. Let say the student likes to bike around the neighbourhood to explore and the student bikes around this very bad neighbourhood with criminals and dangerous people. And the student gets attacked (an extreme example, I know) by a gang member. You now learned physically from the environment that you probably will not be biking or walking this street anytime soon. so you learn from the environment by positive or negative signals.
Disclaimer: This is a very simplified representation, and when you first go into depth, you find so many mathematically functions.
Now there is a lot of math in this book. And I love how a 100 page can contain so much math for the topic of machine learning. And the first few pages he reminds you of the math, which you probably forgot, and I it appreciate that he takes you through that journey.
But after reading this we can simplify machine learning in 7 steps.
I knew how to gather data, and prepare the data. But often I get stuck in choosing a model. In the book there is this picture, like this one :
this picture helped me a lot for choosing a model that fit for me. And he goes on and on about the different algorithm and how it works.
Now this book is a very handy book for looking at all the algorithms, models, mathematically explaining of how machine learning works and so on. And is definitely a must read book. But, but, butt….. I find it sometimes hard to understand the concept. Meaning the book is great, but I am the type of person that want to see how we would code this. I understand the concept now, but I looked forward to how we should have coded this. Because I have made some projects with machine learning but with the help of so many libraries like scikit-learn, numpy etc. It felt like it was easy. The hard part was more on how to code it then the actually math how it works. But overall I will definitely give this book 4/5 stars. I think this meme summarize my problem about the learning of machine learning.