Real-time Learning with Bayesian Bandits

Ted Dunning (MapR)
Data Science, Sutton Center / Sutton South (NY Hilton)
Average rating: ***..
(3.67, 6 ratings)

This talk will describe how real-time learning can be used for advanced A/B testing as well as a variety of targeting problems from online advertising to adaptive document routing to content-based recommendations. The crux of these applications is a family of techniques called Bayesian Bandits. The groundwork of these methods was first published in the 30’s but only recently have the implications become apparent. In contrast with many learning algorithms the key algorithms underlying Bayesian Bandits, notably randomized probability matching can be implemented very simply yet still provide state-of-the-art performance.

This talk will start with very practical demonstrations of the intuitions behind the mathematics, will provide an outline of how to implement them along with pointers to open source code. Then we will provide examples of how to apply these techniques to a variety of practical problems.

While this text will allude to mathematical techniques, it will be approachable for a wide range of audience members from theoreticians to implementors to business stake-holders. The key takeaways for the audience will be an understanding of the intuitions underlying how Bayesian Bandits work and how they can be applied.

Photo of Ted Dunning

Ted Dunning

MapR

Serial startup and artist and open-source innovator, particularly interested in large data systems and statistical modeling.

Sponsors

Sponsorship Opportunities

For information on exhibition and sponsorship opportunities, contact Susan Stewart at sstewart@oreilly.com.

Media Partner Opportunities

For information on trade opportunities contact Kathy Yu at mediapartners
@oreilly.com

Press and Media

For media-related inquiries, contact Maureen Jennings at maureen@oreilly.com

Contact Us

View a complete list of Strata contacts.