Dela med andra

Next generation ad optimisation for global advertisers

Publicerad 2021-12-10

Om uppdragsgivaren
At Bannerflow we help designers and advertisers across the globe to create first-class digital advertising with an impact - at scale. Each month our platform serves more than 6 billion ads, viewed on 500 million unique devices all over the world. This breaks down to a massive 20 000 requests and 1 GB of data transfer - per second, on average, constantly!

Beskrivning av examensarbetet
Our platform allows automatic ad optimization, that the platform will pick the best performing ad from a selection of variants. Traditionally this is achieved by A/B testing. However, by using a multi-armed bandit algorithm, results can be produced faster and thus minimise the overall waste. We have a first version running, using a multi-armed bandit algorithm, but we believe there are huge opportunities for improvement in this area.

Scope of the thesis
The aim of the thesis is to test and adapt multi-armed bandit algorithms with the end goal of reducing waste from serving bad performing ads. Multi-armed bandit algorithms are well understood theoretically. However, in a production setting with different constraints e.g. batching and explainability, where the actual performance and not theoretical optimums are important the situation is not as clear.

The thesis work will include:
- Conducting a pre-study identifying algorithms that should be included based on our specific use case and criterias
- Implement one or several algorithm prototypes
- Tweaking the algorithm parameters with help of historical data
- Testing the prototypes both on historical data and in live operations

Your background
We are looking for one or two students, with an academic background in computer science, machine learning, statistics or similar. We believe you have an interest in applying theoretical knowledge on practical problems.

What´s in it for you?
- Unique opportunity to explore multi-armed bandit algorithms on a huge online advertising dataset
- Work with talented colleagues from around the globe with many years of experience
- Apply your knowledge on a practical problem - your work will become the basis for our next level creative optimization
- Given a suitable algorithm is identified, see the result used in production by our global customer base

Either join us at Bannerflow´s headquarters located at Fridhemsplan in Stockholm, or do the thesis remotely.

Björn Karlström
St Eriksgatan 46
112 34 Stockholm




Sista ansökningdag



Bookmark and Share