Data Science in Practice (Beer Sheva Meetup)

18:00 - 18:30:
Networking and Refreshments
18:30 - 19:15:
Managing Data Science in Retail from Collection to Modeling, Roy Resh, Senior Data Scientist @Trax Retail
19:15 - 20:00:
The Robin Hood Exploration, Shaked Zychlinski, Algorithm Engineer @Taboola

Managing Data Science in Retail from Collection to Modeling

At Trax, we digitize retail data from images. We process it using our image recognition engines and output it in tabular form. While this data is very rich and unique, it also requires quite a bit of preprocessing to make it useful for data science and advanced analytics.
In this session I will present how complicated this preprocessing stage really is and how much algorithmic work is being put into managing and validating the data even before using it. I will demonstrate how we leverage anomaly detection models, and data cleaning techniques to create clean, workable retail data. Lastly, I will show how this data is joined with Point of Sales data to drive real business value for our customers.

Roy Resh, Senior Data Scientist with 5+ years of experience @ Trax. Currently focusing on developing prescriptive models that deliver insights for CPG customers. Prior to that worked as a Team leader and Algorithm developer at Trax in the computer vision group, implementing anomaly detection and data enrichment engines. Has a B.Sc in Physics and Chemistry and an M.Sc in Physics specializing in quantum information from the Hebrew University of Jerusalem. Father to 3 kids, likes doing crossfit, hiking and music.
https://www.linkedin.com/in/alongrubshtein/

The Robin-Hood Exploration

Deep-learning based algorithms are at the core of Taboola's recommendation system, used daily to personalize the reading experience of millions of users world-wide. Unfortunately, DL-based recommendation systems are prone to act like a memorization machine - and repeat the same recommendations over and over again, thus hurting engagement potential. Therefore "Exploration" - enriching the data with recommendations outside of the model is a must. In this talk we'll cover why naive methods of exploration fail and discuss different methods of smart exploration. Finally, we reveal a novel method developed by us specifically for embedding-based deep-learning recommendation systems.

Shaked Zychlinski is an algorithm engineer at Taboola, where he works on deep-learning applications for recommendation systems. He fell in love with this field when he trained a neural-network to play Tic-Tac-Toe, and it pulled a trick on him and won the game. He has a B.Sc and M.Sc in physics, and a constant crave for ice-cream.
https://www.linkedin.com/in/shakedzy/