Amotek HOT topic: Machine learning!
Machine learning, AI, ... These days you hear all kinds of things about them, but what exactly do these terms mean? Our CTO, Filip Van Raemdonck, is happy to explain!
The future is unpredictable. No one knows what is to come. Yet we draw lessons from history and constantly set expectations for our future. We recognise and categorise situations and become suspicious when something does not behave according to our expectations.
Computers, on the other hand, cannot think. They cannot understand situations, they cannot judge, nor can they make decisions based on their feelings. What they can do is help us understand what has happened, what is happening and what will happen. For example, through algorithms that exhibit a form of intelligence, better known as Artificial Intelligence (AI). One component of this is Machine learning (ML), which focuses on algorithms that "learn". This involves training a model with known data to improve its performance.
Suppose we had a company that could store energy, for example hydroelectric energy storage, we would like to know the upcoming energy prices to maximise profits. We know from history that energy prices can change significantly depending on weather conditions or the time of day, for example. We can use machine learning for our business to create an algorithm that can give us a prediction about the upcoming prices.
The first step for building such a model, is to collect the historical data; the training data. These data include, on the one hand, the historical energy prices; the target, what we want to estimate. On the other hand, they include the data that influence energy prices; for example, the weather conditions at the time, the characteristics.
We then examine the impact of each characteristic on energy prices. Using relevant characteristics, we can now train a model that estimates energy prices based on the characteristics.
Algorithms
Within machine learning, there are two different types of learning. On the one hand, there is supervised learning, and on the other, unsupervised learning. When we try to predict a value, as in the example, we speak of regression. Regression is a form of supervised machine learning, where we label data in our training. Here, the goal of our machine learning model is to determine the labels of new (as yet) unlabelled data. Another form of supervised machine learning is classification. An example is handwriting recognition. You can train a model with known inputs, but the number of possible outputs is fixed in this case. The examples below are examples of supervised machine learning models; the algorithm is trained with historical data whose purpose we know.
Besides supervised machine learning, unsupervised machine learning also exists. Here there are two main categories; clustering and anomaly detection. In anomaly detection, we look for patterns in our incoming data. When these deviate too much from the general patterns within this data, we speak of an anomaly. In clustering, data is automatically grouped based on similar properties. A clustering algorithm groups users of a service with similar preferences, allowing users within this group to receive similar recommendations.
