Machine learning — What?
At this point in your life, you have probably heard about, Artificial Intelligence, Machine Learning and maybe something called deep learning, and most of us have had some trouble understanding, well today, let’s talk a little about Machine Learning which is a little part of artificial intelligence, but a really big part of computer science and data, but in a simple way.
Well, let’s try to understand this:
Your computer is really good for making math operations and faster than any human, it can save more information that you can remember and actually in some cases can get some things that we can’t, like waves and all that tech stuff, but would you say that your computer is smarter than you?
Ask for a moment, why you still feel that you are smarter? Do you feel that your computer can’t make complex decisions that us humans can make in no time? Well, it is probably because you are right, you can learn new things on the run… and your computer doesn’t.
But some really smart scientists decide that machines could learn, they realize that if computers are able to learn, they can make better decisions for us every time we ask them to do something. They start to ask, how we can make computers learn? And the most important question, how can we approach computers to operate as the most powerful of all computers? The brain!!!
Well, they decide to make something called “Machine Learning”, they begin to treat computers as Childs and teach them from scratch.
A little scary… right? Well, that is because we have seen many movies about machines destroying the world, but is not like that. At this point in your life, you have a deal with several machines that have to learn many things, the deal is simple, you give them data and they give you experiences, content or even those promotions that you love.
Remember when you are watching movies on Netflix, videos on youtube or listen to music on Spotify, and all of the sudden a suggestion of something you like and didn’t know it was there appear on the screen? Well, some machines just learn that and decide that you could also like that.
Yes, every time you make a click, you watch something, you press the “like” button, take a picture, or sing a paper you are given data to that little baby machine that is learning, and it’s not only you, is millions of people given data in daily basis, and in other cases, universities ask for help from people to get the data of a specific case, for example, photos of people in different moods, to teach a machine to differentiate those moods and then give an application to blind people that will tell them what is the visual mood of the person in front of them.
And actually that sounds either banal or profound, but what if… a machine could learn to identify diseases with just a picture? Imagine if when you get to the doctor you just have to take a picture and some lab and in a matter of seconds your doctor has the help of a machine with millions of data that identify some symptoms that can’t be ignored by the human.
Well, that sounds a litter more fascinating!!! Right???
Ok, let’s talk about how they do that, but in the simplest way possible.
The first thing they do is collect all the data they can, just like when you learn to choose something good at the grocery store, at some point we just buy and buy and buy, and collect that data of what are the good and the bad stuff.
But as you imagined, not all the data is good to learn… like when you try to learn how to buy avocados in the local grocery store and someone sells you a special avocado that only appears one time every 20 years… so the scientists have to clean the data to choose what is better to help the machine to learn.
After the data has been clean, these guys analyze the problem and decide what is the best way to “teach” the machine according to the problem… yes, that sounds weird, but think about it, if you want to learn how to swim, and someone tells you to read a book and stay in your chair because they learn math that way and they succeed, you will think that person is crazy… that’s because the model of learning has to be chosen carefully.
They have 3 options for this, one is called “supervised learning ”, this means that the people in charge of this will give the machine the data and the answers, so it learns whats is what, then they give the machine a set of data without the answer and see if it is capable of finding the answer by itself, yes, just like we did in school.
The second option is called “Unsupervised learning”, that means that the machine doesn’t know anything about the data the give it, but it starts to classify everything by the main characteristics, for example, if we give the machine several photos of hamburgers and hot dogs it will identify the shape and color of all the images and them it will group them in two groups, that way the machine will learn without supervision, that those two items are different.
The third option is called “Reinforcement learning”, in that scenario, the machi doesn’t know anything, so it starts to do everything it can, for example, if you want to teach a dog how to roll, the dog won’t know what “roll” is, but he will start doing everything he can to please you, but when the dog does something like “lay down” you will give him a treat, and when he runs al over the place, you splash a little water on him, that way, you reinforce a “good” behavior and “punish” bad behavior, that, apply to computers, will help the machine to learn something based in what thing get rewarded and what is punished.
At that point, they start to teach the computer applying the model they choose and using the data the selected before, in simpler words, the machine is learning the way the teachers told it was the best way to learn how to solve that problem.
And that’s it… no, it doesn’t stop there.
They are humans, but also scientists or at least very smart people, and they have to check if the data/model they choose is given the results they expect, so they take a bunch of data that they keep checking if the machine has learned correctly.
With this information, they make all the adjustments they need to do, which includes the ability to say… “we didn’t teach it well”, and in that case, all the process has to be repeated.
At the beginning of the concept, this was not something simple, but know it can be done in really simple ways, which is fascinating all thanks to the creativity of the people that have dedicated to this area becoming better, and constantly bend the existing techniques of learning to achieve new and better results.
Finally, the machine has learned to do a specific task and can help many people at the same time.
Has you can see, is no different from when we all learn in school, somebody decides how we were going to learn all the stuff we know now… they even decide what topics we have to learn, at the end of the day, scientists are just doing what we did has children, and that is been a great Breakthru for humankind because we discover something great, doesn’t mean we are doing great things with that.
Now we have the responsibility to create constructive stuff with it, there are thousands of things been done with this amazing technology and some not so good.
For example, some people realize they can make something called “deep fakes”, and that can make a person’s face appear to be saying or doing anything on a video, and it seems so real that in some cases is impossible to tell if it’s real or fake.
But at the same time, there are people working on a way to counter that with the same technology, isn’t that superb?
In the end, we can conclude that machine learning is simple has humans teaching machines how to learn, based on the ways we have learned during of whole life, we are still the most advanced and intelligent computer in the planet earth, and we are working our way to get off those tasks that are keeping us from thinking in new ways to solve complex problems, we are looking at infinite possibilities in which we can get all the knowledge, all the data from a specific topic and get it in one place to help us make better decisions.
I really hope you understand what is machine learning, see you in another post!!!.
Sources
Michie, D., Spiegelhalter, D. J., & Taylor, C. C. (1994). Machine learning. Neural and Statistical Classification, 13(1994), 1–298.
Dietterich, T. G. (1997). Machine-Learning Research. AI Magazine, 18(4), 97. https://doi.org/10.1609/aimag.v18i4.1324
Jordan, M. I., & Mitchell, T. M. (2015). Machine learning: Trends, perspectives, and prospects. Science, 349(6245), 255–260.
