The Mercury Machine Learning Lab is a collaboration between the TU Delft, the University of Amsterdam and Dutch tech company Booking.com. Together they are focused and working on the development and applications of AI for online travel booking and recommendation systems. The beauty of this collaboration is that it provides the researchers with a unique environment where AI ideas can be tested on a real life platform and for the company to get state-of-the-art knowledge on the forefront developments of AI and machine learning.
In Delft the researchers are focused on developing sequential reasoning methods via reinforcement learning. In this type of learning goals are being expressed as ‘rewards’. Combined with sequential learning this can enable machine learning to take good decisions taking into account their effects over a longer period of time, rather than assuming static or “one shot” decisions that are the same every time as in more traditional supervised learning settings. For instance, finding and booking the best possible place to stay is not a "one shot" decision, but the result of a long process of comparing, filtering and searching. A reinforcement learning based recommendation system can mimic the methods of the best travel agents: presenting initial recommendations, listening to feedback on these suggestions, making refined suggestions, listening to feedback etc. 

The fundamental problems the researchers work on at the Mercury lab can have broad applicability once solved. The seemingly different problems of optimally treating patients in a hospital and optimally guiding travelers through the myriad of travel options share at a mathematical level a common problem. The hope is that is that the new reinforcement learning algorithms that will be developed can be validated in the Booking.com environment and find their way to other, harder to experiment with domains.