Our 2019 data science research day

By Gene Ferruzza

Every summer the Data Science team at Valassis conducts 2–3 research projects where interns work with the team and often do much of the heavy lifting. These projects typically conclude in August when we present the results of this research along with other new capabilities developed throughout the year including relevant work performed at local universities and other special interest topics from data science and AI. It has become known as Data Science Research Day.

Along with the Data Science team, this year’s Research Day had attendees from the Product, Engineering, Marketing Science and Operations departments. We also invited some of the data science alumni from Valassis and had eight of our past colleagues join us for the day!

The best way to talk about this event is to run through the topics that unfolded during the day along with the speakers who presented the content. The day kicked off with Ellie Hanna, a Data Scientist on the Valassis team, talking about some of her past cognitive science work on functional MRI (fMRI). The problem that she outlined was our ability to interpret the extreme dimensionality of neuroimaging data comes with a high risk of false positives. She started with MRI images of an Atlantic Salmon and then showed us the various signals her research covered in human brain MRI images based on blood flow and other brain activity.

Next up was one of our interns, Federico Ferrari, a Ph.D. student in Applied Statistics out of Duke. Federico focused his research on improving our solution for “Cold Start” CPA campaigns. This is a common issue when there are no data on past conversions to help predict audiences that will convert in the future. Federico created a new performance measurement metric to help evaluate his work and plans to publish the details of this new technique in a whitepaper. His work is currently scheduled for deployment and will be in operation by the end of the year. To the right you can see Federico flanked by his two mentors Ulric Wong and Saravana Srinivasan.

The keynote speaker was Alexander Volfovsky, Assistant Professor of Statistical Science at Duke University. Alex worked with our Data Science team years ago when we sponsored research at Harvard. Since then he has moved back to NC and has hung his plaque on Duke Campus. He introduced everyone to his work in causal inference and the ability to find barely detectable signals that often are the underlying drivers to human behavior.

Alex will continue working with the Data Science team on AdTech business applications in our data environment. We have found that some of his new methods apply well to answering some of our more challenging questions. Alex and his post-grad team are reviewing projects that we have laid out so we can continue collaboration and innovation of our technology.

The next presentation was provided by our own Susan Xia who has always had an interest in the mathematics behind the latest in cryptology. As Susan puts it, “Cryptography is the art (aka. math) of mangling data into apparent unintelligibility and allowing a secret way of un-mangle it.” Her presentation walked us through the different types of cryptography including Public Key, Lattice-based and ending with Homomorphic Encryption. It was very clear, the harder the mathematics the better the encryption scheme

Ali Taleqani followed with his research results on Smart Parcels. Ali is a PH.D. candidate in computer science, coupling that with post-grad work in finance, logistics and transportation. He came to Valassis from North Dakota State University. Ali’s research at Valassis was focused on sharpening our identification and confidence in location signals that we receive. Ultimately, we want to know if a consumer visits a location and sometimes the location is not well defined, and sometimes the location signal is noisy. Ali approached his solution through both a Naïve Bayes model and a graph theory based model. His work contributes to the accuracy and adds confidence levels to our Graph’s location information. Ali is pictured here flanked by his mentors Schaun Wheeler and Sam Hendley.

Last up for the day was Alex Koryachko, a Data Scientist with Valassis. Alex gave us a thorough look at how natural intelligence compares to machine intelligence. He walked us through where natural intelligence excels, where it is not so good, and where we expect machines to do better. Alex provided a fascinating look at both perspectives mixed with colorful insights and his own natural built-in humor.

We ended the day with a Data Science team dinner joined by many of our old colleagues.