Title: Getting to Continuous Optimization
Most modern analytical systems are descended, in perspective if not in actual code lineage, from "business intelligence" and are, really, something that the software developers who used Crystal Reports to format data pulled from a FoxPro database in 1989 could feel a spiritual kinship to. That is, they are designed to sift through vast amounts of information and present summaries to humans, who then take action (or, less ambitiously, take notes for their next quarterly report).
The current passion for big data, machine learning and predictive analytics has, for the most part, not managed to escape this. When you use supervised learning for churn prediction, the usual goal is to predict who is going to leave the system, and enable humans to take action. Similarly, a typical "A/B testing" scenario usually involves running a small number of simultaneous variations in a system for a short period of time, and then employing a significance test to determine a winner, which then becomes the default behavior of the system.
What if, instead, you want to continuously optimize a complex system? What if you are running an online service and want to customize the game to each player?
How would you do that? In this talk, we'll go through some of the lesserwellknown tools from the machine learning toolkit, and talk about how to design systems that continuously and constantly adapt to the end user.
The examples will be from modern gaming, but the content is more generally applicable.
William “Bill” Grosso has been helping games become more profitable for almost a decade. A seasoned technology veteran, Bill has two decades of experience in building innovative, customer-centric products and delivering high performance software.
Bill is well-known as an expert in both big data and predictive analytics. He has spoken at many industry events, most recently at nucl.ai (http://events.nucl.ai/) (where he has both keynoted (https://vimeo.com/144009041) and presented in the data science track (http://www.slideshare.net/ScientificRevenue/causal-inference-reinforcement-learning-and-continuous-optimization)) andPocketGamer (http://www.pgconnects.com/vancouver/speakers/william-grosso/). You can see videos of some of his talks on our resources page (http://www.scientificrevenue.com/resources/), or on YouTube (https://www.youtube.com/results?search_query=william+grosso+gaming). You can also view many of his presentations on Slideshare (here (http://www.slideshare.net/wgrosso/presentations) and here (http://www.slideshare.net/ScientificRevenue/presentations))
In his previous role as CTO and SVP of Product at Emergent Payments (https://www.emergentpayments.net/) (formerly LiveGamer), Bill provided a scalable micro-transaction platform to leading industry clients such as EA, Sony Online, and Take 2. He’s held executive positions in both engineering and product organizations, been a researcher in artificial intelligence at Stanford University. He has written two books on software design, translated a book on hyperbolic group theory (http://perso.ens-lyon.fr/ghys/articles/groupeshyperboliques-english.pdf), and has authored or co-authored more than 50 scientific publications. Bill has a B.S. in Economics from SUNY StonyBrook and an M.A. in Mathematics from UC Berkeley.