Talk Abstract: Contextual multi-armed bandits for widget optimization.
Mobile applications and websites are growing more complex than ever, with new graphics, functionalities and widgets being added every day. In this ever-growing space it is important to develop new approaches to surface the right content to the right users as many time as possible. While A/B test is a widely used and solid technique, it is not always viable when the number of possible choices is very large, hundreds or thousands of tests would be required to find the best option for each situation. This talk will firstly provide an introduction to the muti-armed bandit problem. Then, a practical comparison between bandits and classic A/B testing will be shown. Closing with a practical Bandit implementation at Skyscanner.
Bio: Marco Bertetti is a Data Scientist at Skyscanner based in London, who has worked both in using reinforcement learning for mobile app content, and in shaping the structure and integrity of Skyscanner’s logging and data. Before joining Skyscanner, he has worked on different problems for a variety of companies ranging from tech startup to big retailer. He obtained a degree in Globalization, International Institutions and Economic Development at the University of Trento before moving to London. In his free time, he likes photography, cooking and rock climbing.