Züri Machine Learning Meetup #4

• Deploying Machine Learning @ Google
Doug Aberdeen (http://research.google.com/pubs/author39485.html), Staff Engineer at Google Zürich

During my PhD I was inspired by examples of machine learning (ML) in games like Backgammon, but I was always disappointed that ML methods didn't seem to have any deeply impactful uses in the real world. In the last 10 years, all that has changed. Companies like Google, Facebook, Amazon, and institutions like Cern, rely heavily on Machine Learning methods to organize data and make it useful. There are hundreds, if not thousands, of instances of ML trained models being used in Google every day. I'll give some insights into how we use ML by covering three productionized examples that I have worked on in the Zurich office: Gmail's spam and abuse detection, Gmail's Priority Inbox, and google.com (http://google.com/)'s ad click prediction.

• The Panini Test
Daniel Stekhoven (https://www.linkedin.com/pub/daniel-stekhoven/60/345/15), Founder of Quantik AG (http://www.quantik.ch/)

Statistics is more than ever a hot topic and still it remains a closed book to the majority of scientists out there. The three basal questions in statistics are; what is a plausible parameter value? Does the data confirm this value? What range of values is compatible with the data? The answers to the first and third question are estimation and confidence interval. The answer to the second question is a hypothesis test - and this is what this talk will be about. Buckle up and get ready for some applied statistics - right in time for the FIFA world cup.
[ slides (http://goo.gl/KLHdJK) ]

• Spark for High-throughput, Scalable, Quantitative Analysis of Genome-Scale Datasets
Kevin Mader (http://www.biomed.ee.ethz.ch/people/maderk), Lecturer at ETH Zürich

Recent improvements in the rate and quality of gene sequencing have resulted in a flood of sequence and marker data. The step of extracting meaning from these datasets has, however, been limited by the ability to analyze these results and compare them with quantitative phenotypes. We demonstrate the application of a new model for Resilient Distributed Datasets present in Spark (http://spark.apache.org/) enable quantitative analysis and real-time exploration of hundreds of phenotypes with millions of samples.

The apéro this time will be offered by Quantik - statistical thinking (http://www.quantik.ch/) - many thanks!