Machines learn and predict future

Stanisław Saganowski, PhD student at Wroclaw University of Technology specialises in such forecasts.

Social network analysis requires many calculations. You need to know - describe, calculate and measure - exactly what happened to the group in the past. Then, researchers build algorithms that allow to create forecasts. Machine learning methods are used to predict the future behaviour of a group. Computers determine the direction of possible changes. This field of research is called the prediction of the dynamics of social groups. What are its practical applications?

"We identify a group, it can be a family that has operator X mobile phones. The second group is the circle of friends of this family, who have operator y mobile phones. By analysing the frequency of calls of particular groups in the past - we try to predict whether any of the groups will want to absorb the other, whether they will change their operator" - said Saganowski.

Reports show that the cost of acquiring new mobile users are ten times higher than the costs of maintaining users in the network. If the operator decides to use the methods proposed by scientists from Wrocław, he will be able to identify two thousand customers who are likely to change the network in a month. This will give him the opportunity to undertake activities to keep those customers in the operator\'s network. It may also help attract customers willing to switch to his network.

Forecasts can also be used in completely different areas, for example to analyse the evolution of diseases. "We have, for example, a group that a decade ago did not respond to 10 types of medicines, and today does not respond 30 types. By analysing the history of changes we can predict whether in the future they become resistant to other drugs or not" - said Saganowski.

Another example is the management of teams in organizations. Forecasting methods may be useful in a company dealing with IT projects, where different programmers often work on various projects. If you completed a hundred tasks - it can be assessed, who has worked best with whom i order to select the best team for the next project. Forecasting methods allow to compose a group of people with optimally distributed competencies and traits that allow them to effectively cooperate with each other.

In his doctoral project Saganowski is developing a set of methods that can be applied in any field, by appropriately selecting the context. The PhD student is one of 500 researchers selected for the Top 500 Innovators programme. In this programme, young innovators travelled to Silicon Valley, Cambridge or Oxford to find out how to commercialise research results and promote knowledge. The qualifying criteria included the candidates\' publications, their motivation, self-presentation skills and knowledge of English.

"I went to Cambridge for two months. Training was focused on our European problems - we discussed Horizon 2020, we considered, in which areas we could compete, how to write proposals. The programme has been completed, but more, this time business programs are planned. We have sent scientists to teach them how to become entrepreneurs. Now we will send entrepreneurs, so that they learn about the world of science. It is a very good direction that will allow us to meet at some stage of commercialisation" - said Saganowski.

He added that an invaluable result of his trip to Cambridge were the contacts he established there with researchers from all over Poland. Top 500 Innovators Association is based on networking.

"Every day we get emails with questions about various problems, experience in proposal writing, or persons to cooperate with. In the case of requests for help there are always some answers. We strongly encourage entrepreneurs to participate in such a program" - emphasised Saganowski.

PAP - Science and Scholarship in Poland, Karolina Duszczyk

kol/ zan/

tr. RL