Surprises when you just start out with Machine Learning

There is no doubt that Machine Learning is a buzzword (or two for that matter) and many people start to dive into it without knowing what eventually will happen to them. They just want to gather any knowledge bit about deep learning and neural networks, not really knowing about the big picture. Like to many others, machine learning was a highly interesting and absorbing topic to me and I just started to learn it, also not knowing what would become of me in a few months. Back then I was taking some big MOOC and secretly dreaming to become a real Machine Learner and doing impossible things like building super intelligent robots and cracking the lotto numbers. But as you advance more in this field, as you speak to professionals, hobbyists and companies you will encounter that Machine Learning is not all there is. It is more of a specialization of something bigger:

Reference: SAS institute

ML can be seen as a part of data science and and a subset of artificial intelligence. But what does that mean for us learners?

Focus on the Fundamentals

Like in any other field, focusing on the fundamentals is the key. When you just start out with programming, you will encounter so many people asking for advice: “Which language is the best? Which technology stack should I use?”

While those are interesting questions later on, especially for beginners the most important question should be: What are the underlying fundamentals out there to explore in my field of interest? E.g. for programming it would be something like the main structures of crafting code, exploration some important well written algorithms and writing clean code. The same fundamental concept might apply to machine learning. As soon as you are done with some courses, tutorials or books, it’s now on up to you to perform. However, you could encounter that having an even solid basis in NLP or deep learning won’t make you a good problem solver. And this is where the fundamentals of data science strike in. If you lack them, you’ll get stuck. So in order to succeed in this field, you don’t only have to focus on where and how to apply which machine learning algorithm, but also it is essential to focus on the fundamentals of data engineering because in the end ML is a specialization of it.

Data Science Fundamentals

This is what I found so far to be the fundamentals of data science which I will definitely need to learn and apply to become a really deep machine learner in future:

  • Conceptual Thinking — Thinking about cases and creating hypothesis to define a problem will be always persistent in your everyday job in the data jungles. To start to train this skill from the beginning on is highly important and underestimated in a world where kaggle and co. lay out their predefined problems.
  • Statistical Thinking — Understand types of statistics and data masses, understand where they come from, how are they to be interpreted, analyzed and processed. All of a sudden you have to ask yourself if this survey is worth working on and occupy yourself with research jargon.
  • Applicationism — No, there is no such word in the dictionary, though, this should be a part of your learning. I like the religious sound of that word, because this is how you should treat practice: religiously. I think that especially in the beginning the ratio should be 50:50 theory vs application. That means: read chapter in a book about data mining? → go mine a data set of your personal interest! Went through a tutorial with visualization of some data? -> Try it yourself with your own data!
  • AI — If Machine Learning is a subset of AI, it would be irresponsible of us to be an ML expert not having a foundation in AI. If you build this one up or at least read yourself into topic you won’t be tempted to solve everything with deep learning like a little TicTacToe game where a fairly simple MinMax Algorithm would be sufficient.
  • OutOfBox Thinking — It is quite easy to rely on what we know or what we did or what others did in similar cases to solve a certain problem. However, a lot of excellent solutions in this world either evolved because somebody came up with a solution that is much easier than anything else before. Or a solution that is super sophisticated on a endlessly high level of abstraction. In both cases people were approaching the problem from a whole different angle and level at which the problem was created. I’m not quite sure yet, how to learn this skill. It’s pretty easy to say, ‘just look at the problem from a different perspective’, but how do you actually do this? I will definitely experiment with that and let you know what I found out.

Each of these fundamentals are worth an own blog post for sure, but I’ll keep it as a rough overview for now. Please let me know if something is of particular interest to you. And don’t be surprised in the future if you transformed from a walking machine learning algorithms collection into a conceptual statistician and an applicational AI researcher with your last name being Einstein all of a sudden. You then acquired all the needed foundations of ML, congratulations! ;)

You are here, so chances are you liked this story :) Click the little heart here on Medium to spread the word and stay up to date about next data goodies. Any feedback is highly welcome! You can also connect with me on Facebook, LinkedIn or Twitter to chatter about snakes and cleanings or just get updated on future material ;)