Applied Machine Learning @ Scale in Seattle - April 2018


Details
• What we'll do
Deep Learning for Fraud Detection
Talk by Ralf - https://www.linkedin.com/in/ralfgunter/
Since our launch in 2011, Sift Science has been detecting fraud by applying modeling approaches like random forests and logistic regression to a carefully curated selection of manually engineered features. While this modeling setup has worked very well overall, it is not as adept as we would like at modeling more complicated end-user behavior, such as that seen from sophisticated fraudsters performing card testing. To fill this gap, we have recently introduced recurrent neural networks into our modeling stack. Since these models naturally operate over sequences, we have found them to be a perfect fit for the task of modeling end-user behavior. In this talk we will walk through our experience applying deep learning techniques to fraud detection, from prototyping through shipping to production.
Amazon Business Customer Acquisition using Machine learning
Talk by Sudheer - https://www.linkedin.com/in/sudheer-ramoji-29588513/
Since the launch of Amazon Business in 2014, business customer acquisition process heavily relied on rule based systems. This process suffered severely both in terms of conversions as well as operational scaling. Starting 2017 we have developed machine learning models to identify potential business customers. This system was later augmented by inference models which exploited the distributions of feature contributions computed by identification models to provide personalized targeting experience. The new ML system significantly improved our performance goals compared to prior systems. In this talk, we will present the over all ML system design and challenges that we had to solve for successful delivery.
The challenges of GDPR-compliance on multi-customer open environment:
Talk by Michel - https://www.linkedin.com/in/michel-goldstein/
In order for Sift Science to be able to accurately detect fraud, we collect and store activity data from our customers. This activity data contains specific information about users and their interaction with our customer's systems. Our customer's systems are different, so is the data we get from each customer. Our machine learning algorithms have successfully allowed us to adapt to this diversity. On the other side, the EU has put forth a new law, GDPR, that requires all companies that process and store person-specific information to allow people to request specific companies to dump, delete and stop processing their data. These requesters should be able to approach any company that does business in the EU and make a request providing a well-defined and limited set of PII to allow them to identify who they are for their GDPR rights. One of the main problems we face for correctly fulfilling those requests is how to provide a consistent set of PII that we request without imposing that same request to our customers. This talk goes through the process that we went in order to build our GDPR compliance strategy.
• What to bring
• Important to know
Food and drinks will be provided.

Applied Machine Learning @ Scale in Seattle - April 2018