Machine Learning

Picture the scenario, you are given the task of teaching a robot to distinguish dogs from cats. At your disposal you have thousands of pictures of dogs and cats. How do you proceed? What instructions do you give the robot to determine what is in the picture? After a period of reflection, it turns out that the problem is more complicated than it seemed.  Most people can handle this task without difficulty, but to solve the problem solving and learning in a way that a human completes the task is a lot more complicated.

One approach is to let the robot learn what is important in the picture to determine which animal is being displayed. This is the essence of machine learning. As the name reveals, it is about a machine that through self-learning understands abstract concepts - such as cat, or the number seven. It may seem that machine learning has a narrow field of application, but there are many application areas. In addition to categorizing abstract concepts, so-called classification, it is possible to train systems to perform forecasts or generate images from instructions in text.

An autonomous system is a system that can independently solve tasks by observing previous examples and thus generalizing a problem. The principle behind autonomous systems is statistical learning, which is used for statistical patterns, which are characteristic of the concepts the machine wants to learn. For example, the number seven is written with a vertical bar that is joined together at an oblique straight line. Since each person has their own writing and inner image of the number seven, the machine examines many examples written by different people and tries to find characteristic features of the concept.It is not always easy to create an abstract concept, such as identifying what is critical for distinguishing dogs and cats, and perhaps not desirable, given the time that must be spent on the task. More efficient with self-learning and this is where machine learning can be a powerful tool.

From the previous example, it appears that data is required to successfully train the system. Depending on the requirements of accuracy and complexity of the problem, the number of data points required increases or decreases. Using enough data is an inevitable factor and can be the difference between an optimal and suboptimal system.  Of course, the more data the better, but it can be costly to collect more data and therefore you usually create more data by manipulating existing data. For example, the number seven should be recognized even if the image is rotated or if colors are adjusted. In this way the amount of data is increased by augmentation without collecting new data points.

In addition to data and a given problem, computer power is needed. The system must learn patterns specifically for the data set, and to find them the must process large amounts of data. If the possibility is not available to invest in a computing computer, there are other options where it is possible to purely computational power from cloud services.

At DING we have experience working with machine learning projects in dynamic pricing, product recognition and object detection. If you have any concerns about using machine learning in your project, do not hesitate to contact us for a consultation. We offer complete solutions, from data collection to a fully functional system adapted to your needs.

click here to read more about Machine Learning

 

Hesam Pakdaman

Head of Machine Learning

DING Design designers

hesam.pakdaman@ding.se