What Is Machine Learning?

During the last few months, I discovered many new concepts around artificial intelligence that I was not really familiar with before joining Artificial Solutions. One of them is Machine Learning. The more I read about it, the more surprised I am about this concept. Let me try to explain what I like about it as I am convinced it will influence our lives greatly in the very near future.

Wikipedia defines Machine Learning as a “subfield of computer science that gives computers the ability to learn without being explicitly programmed”. The idea of this is having a kind of “black box” to which you throw data in and get some data out that solves a specific problem. These algorithms are used for many purposes such as classification, prediction, recommendations or anomaly detection, to give some examples.

Powerful, isn’t it? But, how can this be done? Well, there are three different approaches to do this depending on how autonomous the algorithm is during the learning process:

• Supervised: the algorithm is trained with input and output data that has been mapped with the correct answers. It is like training a child. You teach him some lessons and then you let him think on his own.
• Unsupervised: in this case, labels with correct answers are not provided. It is a much freer approach in which you expect discovering things you never thought. It is useful for example to identify new patterns in a data set or provide some structure.
• Reinforced: in this case you let the algorithm find the answers itself, but you then provide some feedback to reward or punish the behavior. It is a more controlled way to let machines learn on their own.

Depending on the approach that is followed, there are different algorithms that can be used: regression, instance-based, regularization, decision tree, Bayesian, clustering, associative, dimensional, ensemble, to name a few. However, the most interesting for me is the neural network algorithm that is used in deep learning. This type of algorithm tries to simulate the human brain behavior. Explained in a really simple way, it consists of a set of layers in which there are a set of “neurons” that receive an input and send an output signal. Each “neuron” does linear or non-linear transformations and some of them have a more important voice than others. The input signal is transformed as it progresses through the layers until it gets to the final one and voilà, the output is provided. To me, thinking that this is the way our brain works is really impressive, although I guess it might be still an approximation of it…

But it is the application of this technology what fascinates me. Speech recognition, natural language processing, image recognition, spam filtering, fraud detection, text generation, etc. In order to illustrate how powerful this technology is, I selected some examples:

Machines can now put color on black and white images:

Image taken from Richard Zhang, Phillip Isola and Alexei A. Efros.

Even Google writes poems now:

         There is no one else in the world.

         There is no one else in sight.

         They were the only ones who mattered.

         They were the only ones left.

         He had to be with me. She had to be with him.

         I had to do this. I wanted to kill him.

         I started to cry.

         I turned to him.

Read this article from The Guardian for more information about it.

They can also improvise on a Jazz tune:

And can even predict how a static image would became animated:

At this point you might be wondering: How can the algorithm know that the sky is blue, grass is green and a chicken crest is red? Where did it take the inspiration from to write a poem? How did it know which notes to play in a jazz song? How can an algorithm predict movement and create a video?

As you see, possibilities are endless. Thinking of the future can sometimes be scary, but I am convinced that this technology will be a great help to humans and can think of many instances where we can benefit enormously from machine learning.

What do you think?