Looking for new challenges?

Kerstin's Master Thesis: Recommendation system

To navigate the huge amounts of data that are available on the internet, many services use recommendation systems. At Netflix, you are advised about series that are similar to those you previously watched, when shopping at online stores, you get tips of the type “Users who bought this also bought ...“ and hits on Google are sorted differently depending on your previous search history.

I wanted to look more closely at recommendation systems and was specifically curious about what made two items or two users assessed as equal. I developed a content based recommender system in Python with Pandas and compared two similarity calculation methods to see if domain knowledge can affect the system's performance. I got no clear answer, but learned a lot about recommendation systems and became particularly interested in continuing to work with large amounts of data in the future!

Siri's Master Thesis: Neural networks

Training a deep neural network from scratch takes a long time and requires a lot of training data to achieve high accuracy. The project investigated the possibility of using transfer learning to create a classifier who could categorise images on traffic signs. My hypothesis was that if you could retrain a pre-trained network you could achieve high accuracy with shorter training time and less training data.

The pre-trained network was trained on a dataset with over one million images. However, the network had never seen pictures of traffic signs before. To the transfer learning, a dataset with only a few thousand images was used. My results showed that it worked well to use transmission learning to achieve relatively high accuracy with little amount of training data, and that performance can be greatly improved by setting up the network's hyper parameters well.

Summer project 2018: Tiny Cosmo

Tiny Cosmo is a mobile space game built in Unity and the result of the summer project 2018. A team was assembled by three people with different experiences, backgrounds and interests. After six weeks of intensive, fun, educational and challenging work, we were able to present a complete version that exceeded all our expectations.

Everyone was involved in the different parts of development and curiosity was the team’s incentive. The work included everything from design and creation of game music and game trailers to the set of database, web development, game programming and launch of the app. In addition to learning to develop a game from scratch, the purpose was also to build a good project structure, and we learned a lot about planning, group dynamics and team building.

Sara's Master Thesis

Web services with users all over the world generate traffic that travels long distances if the application's servers are in remote locations. Now that cloud services are supplied by companies with data centers scattered around the world, there may be a possibility of utilising these to reduce the distances the traffic needs to travel.

The goal of the thesis has been to investigate whether it is possible to develop a system that automatically can start and shut down servers on the data centers that are geographically closest to the currently active users. The use of a service may follow a circadian rhythm, whereupon it is desirable to have the most servers in the places where, for example, it is night time.

The study showed that the application used for testing could not benefit from the system. However, it is possible that other types of applications could benefit more from this system.