Skip to main content

Transfer Learning - Getting the Most Out of the Data You Have, Not the Data You Want

Brian Dalessandro (Capital One)
Hardcore Data Science Gramercy Suite
Average rating: ****.
(4.17, 6 ratings)
Slides:   1-PPTX 

A basic tenet of building predictive models is that the data used to build the model should have the same distribution and correlated relationships as the data being used in both validation and the application. There are many instances however will this rule must be broken in order to build a feasible predictive modeling solution. Common examples of this include the ‘Cold Start Problem,’ ‘Mass Customization’ and modeling off of rare events. In all of these cases, getting enough of ‘the right’ data is either impossible or prohibitively expensive. In this talk we will cover the strategy of Transfer Learning, which is the sub-field of Machine Learning that involves using knowledge learned from auxiliary data sets to improve the predictive power of the actual problem at hand.

Most real-world applications of predictive modeling include, or could benefit from, some application of Transfer Learning. This talk will cover the basic strategies of Transfer Learning and show under what data and modeling conditions they apply. Also, we will present some specific methods that have been proven effective in the applications of online advertising and spam detection.

Photo of Brian Dalessandro

Brian Dalessandro

Capital One

Brian is a practicing data scientist with 12 years of modeling experience, ranging from high energy particle physics to online advertising. Brian is currently VP of Data Science at M6D, an online advertising firm. Over the past 4 years, Brian has been a lead researcher in developing media6degree’s patent pending machine learning technology. Brian’s current R&D interests include building autonomous machine learning systems over big data architectures, causal inference and transfer learning. Brian has published multiple papers on these topics in top academic journals and conferences. Brian is also a co-chair of the annual KDD Cup Data Mining competition.

Prior to joining media6degrees, Brian was a Senior Research Analyst at and a risk modeler for American Express. Brian holds an MBA with a concentration in Statistics from NYU and a BS in both Mathematics and French Literature from Rutgers University.


Sponsorship Opportunities

For exhibition and sponsorship opportunities, contact Susan Stewart at

Media Partner Opportunities

For information on trade opportunities with O'Reilly conferences email mediapartners

Press & Media

For media-related inquiries, contact Maureen Jennings at

Contact Us

View a complete list of Strata + Hadoop World 2013 contacts