Taking Machine Learning to Production: A View From the Trenches
While experimenting with Machine Learning models in lab environments is key to the success of new predictive products, exporting these models to a 'real life' production environment usually reveals many unforeseen challenges. For eBay, a company that connects millions of buyers and sellers around the world daily, overcoming these challenges is crucial to building successful new products. As the company works on its transformation program to improve its platform for buyers and sellers, they are investing in predictive technology in order to build the world’s most comprehensive product catalog and pricing guide.
In this talk, different subjects that are relevant to the topic will be discussed using 'real-life' examples from predictive models in which eBay is currently investing. For example: a) finding a good fit between the business challenges and the optimization problem; b) when is it necessary to interpret the model and what are the best practices for doing so; c) how to integrate domain expertise into our models; and d) how can we use concepts from Transfer Learning to overcome the training / serving skew.
Yoni Acriche is a lead data scientist at eBay where he is leading the research and algorithm design of the company's strategic initiatives on price-demand prediction. This initiative aims to improve the buyers and sellers understanding of the price differentiating attributes. Prior to eBay, Yoni was the head of data science at Salespredict (acquired by eBay)-- a startup that designed the next generation predictive sales solutions to transform the way companies acquire and retain customers. At Salespredict, he led the data science team’s vision of "Automatic Data Science": an on-going effort to create a product that is completely automated without the need for an expert in the loop. He focused on predictive data mining and the construction of algorithms that leverage information found on the web as well as external dynamics to predict sales conversions. Yoni is also a contributor to the Harvard Business Review, where he writes about the intersection of business and predictive technology. Yoni holds an M.A in Statistics and a B.A in Economics, both from The Hebrew University of Jerusalem.