Machine Learning (…for grandmas)
¿What is machine learning?
Every day we are in contact with telephones, tablets, computers, a lot of gadgets that make our lives easier with all the amazing things they can do but are really “dumb machines”. These devices are very good at obeying orders: If you tell it to show you the photos when you click on the gallery icon, it will; if you tell it to sound the alarm at a certain time, it will, and effectively the screen will slide as indicated by the movement of your fingers. Nothing occurs to itself. You always have to tell them what to do.
And that’s not all, before those computers reach your hands, a programmer must tell them how to do each of those tasks, step by step; this makes its capacity limited because there are many things that humans do not know how to do, and others who know how to do but do not know how to express in the form of instructions understandable to the machine.
Things have started to change; Great minds have found a way to teach computers how to learn on their own and that is what we call Machine Learning. Thanks to machine learning we can now achieve such amazing things as making cars self-driving, or automatically receiving suggestions for products that might interest us when entering an online store.
The way we teach computers is by showing them lots of sample questions and answers (dataset), without telling them how the right answers can be found. Using the examples, the computer tries to find a way to guess the correct answer. The computer searches for similar questions you have seen before and uses the correct answer seen above to tries to learn how to make the best guess possible based on some methods (we’ll see some later).
Supervised learning method
Here we find one of the clearest definitions for supervised learning:
“In supervised learning, the computer is provided with example inputs that are labeled with their desired outputs. The purpose of this method is for the algorithm to be able to “learn” by comparing its actual output with the “taught” outputs to find errors, and modify the model accordingly…
…For example, with supervised learning, an algorithm may be fed data with images of sharks labeled as fish and images of oceans labeled as water. By being trained on this data, the supervised learning algorithm should be able to later identify unlabeled shark images as fish and unlabeled ocean images as water.”
Let’s learn a little about unsupervised learning with the following example:
Suppose we have the following three instances, also called data points or observations.
Now we are going to represent those previous instances through a data table:
In the first column we have a unique ID that helps us identify each of our observations, each of which is represented by a row.
The second is the correct answer. It is what we would like the computer to learn to identify when we show it an image like this.
In the following columns are the Pixel Colors, every time you look at a digital image, that’s you making sense of something that’s stored as a bunch of numbers. These numbers correspond to the chromatic representation of the colors with RGB (Red — Green — Blue) and they are inputs that the computer will learn from.
With this information, the computer can use a machine learning algorithm, which is simply a recipe that uses this information as ingredients to recognize new instances, next with our example, it will be able to identify fish, turtle and oceans.
Unsupervised learning
Unsupervised learning takes place when we don’t have the data tagged for machine training by comparison, as we saw in the previous example.
Now we are going to take some untagged instances:
These images are not tagged, now try to find a characteristic or criterion to separate these two images into two groups. Could take into account aspects such as color, size, number of levels, without or with garage, there are multiple possibilities. if we group it by number of levels, for example, two possible labels to distinguish the groups would be: two-story house and one-story house, in unsupervised language the labels could be a little more generic, like group 1 and group 2. Since this language does not need to group in a language that is meaningful to the human being.
Unsupervised learning helps you find inspiration in the data by bundling similar things together for you, this is called clustering. But is not magic, not having correct answers to compare against, the grouping result may or may not be what you expect, there will be similarities or patterns depending on the algorithm you define for it (the instructions you give the machine to organize the data). There are many different ways to define similarity, you must try different algorithms until you find the one that works best for you.
Another important factor is the amount of data, as more data is available, the ability to make better decisions increases, something similar to what happens to humans in real life, the longer you live the wiser you will be for all the experiences that are accumulated with the weather.
Human Bias
There is a human bias in machine learning, from creating an algorithm to interpreting the data, and until now almost no one has attempted to solve this big problem. When our bias is responsible for selecting a training rule that shapes the creation of a machine learning model, we are not creating artificial intelligence: we are only forcing the learning model with our own flawed observations.
I will take as a reference a very clear example that is in this article and that I invite you to visit if you want to delve into the topic of biases in machine learning:
The classic example used to describe this bias is a machine learning model that’s designed to differentiate between men and women in pictures. The training data contains more pictures of women in kitchens than men in kitchens, or more pictures of men coding than women, then the algorithm is trained to make incorrect inferences about the gender of people engaged in those activities due to prejudices that occur in the real world, represented in the data.
Clearing the Confusion
Often the terms Machine Learning (ML), Artificial intelligence(AI) and Deep Learning (DL) are used interchangeably, but the truth is that they don’t refer to the same thing.
Artificial Intelligence refers to the incorporation of human intelligence into machines. For example, such machines can move and manipulate objects, recognize if someone is moving, or solve other complex problems.
The Machine learning you learned about in this article is a subset of artificial intelligence, necessary for machines to learn from data and take advantage of the patterns it discovers through them to make decisions.
Deep Learning is a subset of ML; it is a technique for machine learning. DL is the next evolution of machine learning. DL tries to emulate the human reasoning process, every time we receive new information, the brain tries to compare it to a known element before making sense of it, which is the same concept that deep learning algorithms use, through a model called neural networks.
Finally… Machine learning Approaches
Currently there are many applications of Machine Learning in our daily life:
- Netflix is a clear example, it takes advantage of key information from its users to have a high level of customer retention, captures data related to the time and days in which you view the content, the type of series you choose, the time when you pause or go back the series, all this to give you the best recommendations of the movies and series.
- Fraud detection is one of the most necessary applications of machine learning. The number of transactions has increased due to a large number of payment channels: such as credit / debit cards and smartphones. Every time a customer makes a transaction, the Machine Learning model takes a complete x-ray of their profile looking for suspicious patterns.
- Another very popular application of machine learning is dynamic and automatic pricing, such as the Uber-determined board for each trip that depends on factors such as traffic, demand and traffic in the area, the destination, the prices of competitors, etc.
- Commonly we use Google Maps, an application that allows us, thanks to Machine Learning, to know the route that is most convenient to reach a certain destination, the one that takes the least time to travel, not only because of the distance but because of the amount of traffic that may be Present
- Social networks, like Facebook, collect information about our activity on the site and information from our profile to suggest people we may know and content that may interest us.
Thanks for reading to the end. I invite you to consult some of the sources on which this article was based to discover much more about Machine Learning:
