How Learn Machine Learning And Become the Best? Find Out.
Humans created machines to free themselves from the drudgery of work, and this was a wise move. Machines have helped humans immensely by making their lives easier and less physically strenuous. But at some point, it was noted that machines can not only do physical labor but can also think, in fact, they might be better thinkers than humans because they lack emotion or prejudice.
Humanity is now in a dilemma, for machines are taking over more work than humans can handle and at the same time make humanity lazy. The solution to this problem is to create even smarter machines that will do all of the work so humanity, which will be relieved of hard labor, will be able to relax.
Machines can do human work better than humans, so they will eventually take over most of our jobs and the only job left for us will be to create more intelligent machines.
So it is with ‘becoming a machine learning engineer’: if you are willing to work hard, and really think about abstract problems like this flowchart problem I had to solve on punch cards 30 years ago, then you can become a machine learning engineer too.
But if you are not willing to slog through these kinds of problems, then perhaps machine learning is not the field for you.
Becoming a machine learning engineer is difficult, and you have to think very abstractly like other engineers do. For example, way back in the day when we were using punch cards to program computers (no joke) I had to write my programs in an obscure language called ‘Fortran’ that was not at all natural for me.
Then, I had to laboriously write out on paper a ‘flow chart’ like the one that follows here:
- This is a flow chart of the algorithm I devised to solve this problem. This flowchart was nearly 1 inch wide and 3 inches long! The computer had 256 punch cards, each one 8 inches long by 4 inches high; there were only 16 columns on each card.
- Nowadays, you can solve exactly the same problem by writing four lines of code in R or Python.
How do I start learning machine learning?
To learn machine learning, we need to understand the basic concepts and ideas behind it. We have to build up our knowledge by studying mathematics and computer science, ideally starting from first principles. Once you get a solid grounding in these subjects, you are ready for practical applications of machine learning.
Then you must be able to apply the concepts and ideas behind machine learning. You need to write code, understand programming languages like python, C++ or Java. Then you need a software framework that can do all of this for you so that your time is spent writing program logic rather than low-level coding.
Learning machine learning means, in practice, creating a classifier. We create a classifier that can take input and output an answer based on this input. The ideal situation would be to have the computer program learn from examples of correct answers so that it can do everything by itself.
Machine learning is also based on statistics. Therefore, it would be very useful to have a good grounding in statistical concepts.
Machine learning is related to artificial intelligence. With machine learning, we create a computer program that can do tasks for us.
Humans have only just begun to really understand the machine learning that drives their world, so I’m sure it will be some time before they come to fully appreciate its power.
The two main classes of machine learning are unsupervised and supervised.
Unsupervised is when the computer figures out stuff on its own, with no human help. Supervised is where you tell it what to look for and then reward it when it finds those things.
Unsupervised is older. It has been around for decades, but it is not yet very good at anything. Supervised machine learning on the other hand has become ridiculously good, so much that it can be used to recognize what you are looking at in a photo and make accurate predictions about your preferences.
The unsupervised learning type is better at dealing with the unknown, but it requires a lot of time and data. The supervised version is more accurate for known subjects, but the amount of data required to train it can be ridiculous.