What on earth is machine learning?

If you think machine learning is just about teaching robots how to do stuff for you, like picking up your dirty laundry and cooking you breakfast, this is the article you’ve been waiting for — but there’s so much more. Basically, machine learning consists of training learning models so that they can later predict and make inferences. Now, I’ve used a few terms that you might not be familiar with (model, training, and inferences) but don’t worry — I’ll explain it all below.

Model: An AI (Artificial Intelligence) or imaginary machine with an entrance and exit which you give certain information. And from that given data it will be able to make deductions, classify things, automate some processes, and more giving you a resulting data. There are many things you can do with a model — it all depends on how you want to use it.

Training: giving your model information to analyze. After doing so, the model can begin processing new information, and from there it can start automating, classifying, etc.

Inferences: using the trained model to make a deduction based on a set of characteristics, and find the best solution. For example, finding the survival rate of people who’ve come in close contact with a particular disease.

One of our developers with experience in machine learning, José, explains:

“You have to give the ML model previous data so that it ​​can process it and thus give you the pattern you need to group future information. For example, if you wanted a model to differentiate men and women, you would train it by sending different images of men indicating their gender and the same with images of women. After a while, you will be able to calculate what dimensions a face should have in order to be considered as a man or a woman.”

There are many ways to train models using different methods, called algorithms. The algorithm you choose will depend on the data that you want to use and how you want to process it. An algorithm is a process or set of rules to be followed in calculations or other problem-solving operations, especially by a computer.

You don’t necessarily have to create new algorithms, you can use existing ones, published by everyone on the internet, and make a few tweaks. Usually, what is done is cross-training, which basically means grabbing the same data and training a model using a specific algorithm, then training another model with the same data using a different algorithm and see how well, and accurate, they both go.

“Technically, you make both algorithms compete against each other to see which is the best that gives you the most precise data, then keep that one.”

Machine Learning helps us as human beings because it gives us the ability to process huge amounts of information that we couldn’t possibly process manually. As great as a dirty-laundry-robot would be, I hope this overview has helped you see that machine learning can be so much more.

Powered by Tekton Labs