Enabling Deep Learning in Real-Time Bidding
Machine learning and AI have been applied extensively in many tasks of computational advertising, like user profiling and yield management. Models are getting more and more complex for these tasks in recent years. However, for bidding optimisation in Real-Time Bidding, it is still hard to apply complex models because a bidder is required to return a prediction in a few ms with thousands of QPS workload. In this talk, I briefly introduce MediaGamma’s end-to-end implementation of the bidding engine. Besides loading a Neural Network model, the process includes a few more challenges like guaranteeing the feature consistency, augmenting bid requests with additional features, and keeping tracking performance of each iterations.
Dr. Shuai Yuan is the VP Data Science of MediaGamma. He supervises and works with the team to build end-to-end data model pipelines to solve challenging problems. He joined the company in 2014 after gaining his Ph.D. in University College London. He has worked with a number of companies such as AppNexus and Bright (acquired by Linkedin) on various topics in computational advertising, including bidding algorithms in RTB and floor price optimisation. He has published several papers and co-hosted the tutorials on RTB in top-tier venues including CIKM, SIGKDD, WSDM, ECIR, and ADKDD. He also published the first empirical study on RTB auctions.