What is Machine Learning

You have probably read the term Machine Learning in the news. It is the new buzzword. Many companies are starting to apply it in their business and their projects. It seems this new technology has come to stay. What exactly is ML? I will try to explain it at a high level in this post so you can have an idea of its numerous applications.

Machine Learning

The objective of Machine Learning is to create algorithms that can learn from data. The idea is to try different mathematical combinations to decipher the relation that different variables have.

The main aim is to develop programs that can automate complex tasks. For example: Create a program that automatically differentiates if an image belongs to a cat or a dog. This may not seem a very impressive task, but was one of the focuses of the field for many years.

Machine Learning can be divided in a subset depending on the type of data with which we are working.

Supervised Learning

What are we referring to when we talk about supervised learning? This are the algorithms that must be trained using labeled data. The previous task of cat vs dog would be an example of supervised learning. The data provided to the algorithm should be images labeled as cat or dog. The main idea is that if we train a cat detector using images of many different cats, when we give it a new image that it has never seen, the algorithm will be able to classify it correctly.

We have been talking about classification, but Supervised Learning also includes regression. The difference between classification and regression is the type of prediction. In classification we predict a categorical variables, e.g. cat or dog. In regression we predict a numerical value, e.g. a stock value.

Unsupervised Learning

In contrast to Supervised, Unsupervised Learning data do not have a label. The objective is not to predict a value, but to find the relation between existing data to discover anything that connects them.

Some examples of Unsupervised Learning would be:

• Clustering → Here we start with some unlabeled data which we want to separate into group of similar characteristics. E.g. We have information such as age and anual salary. If we work a company that produces fridges and wanted to create a campaign to advertise our new disruptive fridge. We would want to put more focus en people with a higher purchasing power. Here is where a technique such as clustering would enter.

In the graph we can see that we would have 2 groups or clusters. One containing young people with low purchase power another with people older than 35 with higher purchasing power. When paying for an Ads campaign we’d probably prefer to put focus on the second group.

Another field of unsupervised learning would be Anomaly Detection which has direct applications in use cases such as fraud detection. Dimensionality Reduction is another field used when we have data with many variables, and want to reduce its size, keeping only variables that can provide us with the most value.

Reinforcement Learning

This is one of the fields in the Machine Learning ecosystem that has gained more fame recently. At a high level it consists on training an agent in a simulated environment using reward functions. With this we aim to reward the agent when it does a good job, and punish when it does a bad job. Its more direct application is video games, since it is easier to train an agent in a simulated environment.

In 2016, AlphaGo, un agent created by Google’s Deepmind, won against one of the best players of the world in the Go game. This game is considered of extreme computational complexity since it has an incredible number of possible moves in each turn.

Computer Vision

The Computer Vision field includes all those tasks related to images and videos. Its main objective is to automate all the jobs that people can do almost perfectly regarding sight. E.g. Detect with just a glance the content of an image or give a description of it.

The example I mentioned previously of differentiating between cats and dogs would belong to the image classification subfield.

Computer Vision doesn’t belong directly to the Machine Learning field, but it inherits techniques from it. This is because Machine Learning has allowed this field to evolve drastically fast in the last years. This acceleration has been greatly caused by the use of Neural Networks and the field of Deep Learning.

Natural Language Processing

Just as in Computer Vision the NLP field inherits techniques from Machine Learning.

This field investigates all those tasks regarding text and its understanding, such as text translation, topic modeling or the development of chatbots.

Conclusions

In this post we have talked about the different fields in Machine Learning and its possible applications. We haven’t entered in this concepts in detail, I’ll do this in future articles.

I hope you have enjoyed this post. If you have any doubt don’t hesitate to ask.

If you want to read my blog in spanish: themachinelearners.com