Deriving patterns from data by enabling the computer to learn by itself


Machine learning is a buzzword with good reason. Enabling the computer to learn to derive patterns from the data provides us with a new way of learning from the past, and making better decisions in the future. To be able to understand what machine learning really is and what it can mean for your organization, one can best think in two different worlds.

The world of data aims to be as good a copy of all measure things in the world we live in. A big advantage of storing a lot of data is that we can view the world from different perspectives. This enables us to draw connections without prejudice and discover new relationships.

Data by itself is actually not really very interesting. A logical first step is to visualize the statistics to be able to derive lessons from it. Machine Learning (ML), or Artificial Intelligence (AI), offers great opportunities for the next phase. It enables us to create connections within the data that we are not able to observe by ourselves.

The computer is used to analyse historical data with a defined goal. It sort of develops the software to make for example a weather prediction by itself. This basically works the same as how we ourselves make predictions. Based on the month and the place we are at, we can make a reasonable prediction of what the weather will be like. The computer does the same, but is able to take way more examples and variables into account (e.g. distance to the sun and presence of clouds).

In summary, Machine Learning provides organizations the opportunity to exploit data from the past to discover patterns that may escape us, and to tell something about what is likely to happen in the future.


The ability to make automated decisions based on images of complex situations, and the power to predict the content of an image, is something that has only really come up in the past few years. The Machine Learning specialization called Deep Learning, makes this possible.

When thinking of data, people often imagine files with rows and columns and cells that contain something. Images are basically just a collection of pixels. While images are definitely a type of data, they are a very unstructured type of data. Making decisions based on images is therefore highly complex. The rise of Convolutional Neural Networks makes this possible. These are able to discover patterns in images, enabling them to for example distinguish images of cats from images of dogs.

Object detection neural networks take this to another level. Not only are they able to classify an image. They are able to predict the exact coordinates of objects within the image, and then classify the type of object it is. In this example we are predicting plastic pollution in rivers. This is highly valuable as because of this technique, factual insights are collected as to how much plastic is in the river.

Follow us:

© 2020 by Soda science