Large-Scale Machine Learning For Display Advertising
Abstract: In display advertising, we must decide which banner ad to show to which person at which time, in real-time for hundreds of millions of users. Predicting future buyers is a good strategy, but marketers also wish to expand their reach to new audiences, different from previous buyers. In this talk we begin by giving an overview of the display advertising problem in Akamai’s Advertising Decision Systems (ADS) group, and some of the machine learning problems that arise from it. We then present a new algorithm for training linear Support Vector Machines over such large datasets. Our algorithm assumes that the dataset is partitioned over several nodes on a cluster and performs a distributed block minimization along with the subsequent line search. The communication complexity of our algorithm is independent of the number of training examples. With our Map-Reduce/Hadoop implementation of this algorithm the accurate training of SVMs over datasets of tens of millions of examples takes less than 11 minutes.
Bios: Rosie Jones is Director of Computational Advertising at Akamai Technologies. Her research interests include computational advertising, web search, geographic information retrieval, and natural language processing. She received her PhD from the School of Computer Science at Carnegie Mellon University under the supervision of Tom Mitchell, where her doctoral thesis was titled Learning to Extract Entities from Labeled and Unlabeled Text. Prior to Akamai, Dr. Jones was a senior research scientist at Yahoo! Labs, where she worked on sponsored search advertising and query session analysis. She has served on the Senior PC for SIGIR and is a Senior Member of the ACM.
Dmitry Pechyony is a Research Scientist at Akamai Technologies. His research interests include statistical and computational machine learning, computational advertising and data mining. He received his PhD from the Technion – Israel Institute of Technology under the supervision of Ran El-Yaniv. Dmitry’s PhD thesis dealt with theoretical and empirical aspects of transductive learning. Before coming to Akamai he held a postdoctoral position at NEC Labs, under the supervision of Vladimir Vapnik. Dmitry has served on program committees of many recent machine learning and web data mining conferences.