The topic of the third event of London Computational Advertising & Behaviour Targeting Group will be about building large-scale machine learning models with highly efficient architecture for processing and analysing big data in performance-driven Real-Time Bidding (RTB) display advertising.
RTB enables the advertisers to perform impression-level evaluation and media buying, which turn out to be a machine learning problem supported by fast and scalable system to processing and analysing large amounts of data. Thus the demand-side platforms (DSPs), which help the advertisers perform intelligent impression-level bidding in real-time, is a perfect practice of such big-data machine learning systems.
With the scalability and efficiency considered, there is a trade-off between the model complexity and accuracy. For example, for the user click-through rate (CTR) prediction module, logistic regression provide slightly lower accuracy but much higher efficiency than support vector machines (SVMs). In such case, within the same training time, more data instances and features can be fed into logistic regression than SVM. As a result, logistic regression indeed provide a higher CTR prediction accuracy than SVM.
We’re delighted to have two very experienced speakers from two DSP companies to talk about their research and practice in building big data models and architectures for the high-performance bidding engines in programmatic media buying.
Dr. Edward Snelson, Lead Data Scientist at Adform
Adform is a leading buy and sell side AdTech platform, providing real-time bidding solutions, ad-serving, data-management, and many other functions. In this talk I will outline some of the projects we are involved in at Adform London Research, the real-time data pipeline architectures we are working with, and modelling approaches we use for predictive analytics and optimization.
Ed studied physics for undergraduate degree, before completing a PhD in machine-learning at the Gatsby Unit, UCL. He went on to Microsoft Research Cambridge to do a post-doc in information retrieval, followed by a stint in the world of finance, then back to Microsoft where he worked for Bing Search Ads, optimizing the parameters of the ad-auctions. In May 2014 Ed joined Adform to build a team of data scientists here in London.
Dr. Richard Hui Li, Senior Data Strategist and Analytics Lead UK/Europe, DataXu
DataXu was founded by MIT rocket scientists who wrote the combinatorial language that guided NASA’s Mars mission plans. These scientists – joined by co-founders with extensive digital media expertise – created an industry's first real-time multivariate decision system to bring science to the art of marketing. In this talk I will outline DataXu's Digital Marketing Cloud platform, with a special focus on the Active Analytics(TM) technology stack. I will explain how such technologies leverage big data and analytics to make faster and smarter decisions to improve marketing ROI.
Richard received his PhD in Computer Science from Leiden University, the Netherlands. He worked as a research staff member in SAP Corporate Research, and was one of the core members who developed and commercialized the record-breaking VectorWise database engine. He was recently a Product Manager in Analytics Innovation at Pega, managing its Predictive Analytics products and Big Data platform. Currently he is leading the Client Analytics Team for UK and Europe at DataXu.
Venue: 1ES 5th floor reception
Date: 15th April 2015
18h30 - 18h40 Weinan gives a welcome talk, introduce the group, guests and thank sponsors and state the rest of the proceedings.
18h40 - 19h20 Edward Snelson’s talk
19h20 - 19h50 20 min break - beer and pizza served
19h50 - 20h30 Richard Hui Li’s talk
20h30+ Drinks and networking.