Description
PyData London 2016
With an inventory of millions of products, finding duplicates is intractable by hand. Neural networks can be used to learn a representation that maps duplicates close together in a manifold.
We trained a network to perform multi-task classification. Using this as a basis, we trained a manifold that maps images to a space where duplicates have a small distance and different products have a large distance.
Slides available here: http://www.slideshare.net/CalvinGiles/finding-needles-in-haystacks-with-deep-neural-networks