Machine Learning


This is the second article in the series entitled “What is AI?” If you have not already read the previous article, we recommend that you do so. To access it, click here.

Today AI has taken the form of algorithms in Machine Learning (ML). An algorithm is a function or model that a computer program can follow and run. Previously it was often a programmer who created an algorithm with rules and conditions, which the program then followed blindly. The idea behind Machine Learning is that the algorithm itself learns to continuously seek out patterns and rules. This is a must as the sheer quantity of data is enormous, with rules and patterns becoming too complex for human beings to interpret.

But how can a computer program learn things that a human being can? In the same way as a human being does, the algorithm must be trained to handle a specific task within certain given parameters. All this learning based on accumulated experience has to be saved somewhere. Human beings save this information in the brain, which is a complex network of neurons and synapses. When a human being registers an impression, such as a picture, information about this picture is sent through the brain. If it is something the person has seen before, there are probably already strong synapses (links) that cause certain neurons to come to the fore (we understand). A simplified example of this is that we find it easy to pick out a family member in a large group of people, since we have often seen this family member on previous occasions (we have trained a strong link in our mental network) thus generating a clear indication. This phenomenon is known as the “grandmother neuron”.

One group of algorithms within ML is known as the Artificial Neural Network (ANN) and it tries to resemble a real neural network as described above. It does so by simulating neurons such as nodes and synapses in the form of weighted connectors. Just like with a human being, you have to train the network by means of a lot of input data so that it can properly update the weighted connectors (in other words, so it can “understand”).

In simple terms, the general learning process can be in one of three ways:
1. Unsupervised learning: You find patterns and rules by yourself, without outside interference.
2. Supervised learning: You learn from someone else.
3. Reinforcement learning: You have a goal for which you aim, and test various methods of reaching that goal, receiving a reward that varies with your performance.

In the following three blog articles we will look more in-depth at these different ways of learning, so keep your eyes open for upcoming articles.

Read about unsupervised learning.