Fashion DNA: Structural Feature Mapping in the World of Retail

Image source: Zalando Research

When applying artificial intelligence to the world of fashion, in which the clothing and other articles involved have many varying individual properties, a meaningful structure must be defined.It seems natural to define it in terms of similarity of items: every item then has it's well defined location in an abstract space, with similar items being close by. At Zalando Research, they've created Fashion DNA to make the properties of their products more accessible, by collecting disjointed information in their catalog and mapping it into an abstract mathematical space - the "fashion space". There, the item is represented by a vector, or it's "DNA", and is used to bring order to a chaotic collection of fashion products.Roland Vollgraf, Research Lead at Zalando Research, will join us at the Machine Intelligence Summit in Amsterdam to share expertise in this area. I spoke to him ahead of the summit on 28-29 June to learn more.Can you describe a short overview of your work at Zalando? I am Research Lead in Zalando Research. I lead a team of 15 Research Scientists in the area of Machine Learning and Artificial Intelligence.   What do you feel are the leading factors enabling recent advancements and uptake of machine learning in the fashion industry? It is clearly the vast amount of data that is generated in the online retail business, paired with continuously increasing computation power (today mainly provided by modern graphic processors).   What present or potential future applications of Fashion DNA excites you most? Our Fashion DNA is a mapping of fashion items into a mathematically convenient Euclidean space. As such it serves the purpose of a general indexing of our assortment for various applications like search or recommendations. However, the mapping is not uniform. I  personally find it most exciting to see what's at the white spots in the feature space, i.e. at points which none of our current articles directly map onto. Which hypothetical new shoes or dresses could be found there?   Which industries do you feel will be most disrupted by machine learning in general, in the future? In my opinion it's self driving cars that will be the most disruptive AI technology in the near future. In the consequence, this could make it fully obsolete to drive and own a car yourself - a scenario that would completely change the sight of our cities, I believe.    What developments can we expect to see in the fashion industry in the next 5 years? Among others, there will be completely new customer facing applications on online retail. Think of better and smarter search experience, like conversational search allowing to ask for inspiration while entering a dialog with the system. Advances in computer vision and deep learning for image manipulation techniques will allow for virtual try-on applications in a way that a selfie made by a user can be realistically augmented with any type of apparel she is about to shop online.

If you’re keen to learn more about the advances of deep learning and their impact on business and society, join RE•WORK at the Deep Learning Summit in San Francisco this 25 & 26 January.

Confirmed speakers include: Andrej Karpathy, Director of AI at Tesla - Yves Raimond, Director of Machine Learning at Netflix - Peter Carr, Research Engineer at Disney Research - Yannis Agiomyrgiannakis, Senior Research Scientist at Google - Eli David, CTO at Deep Instinct - Andrew Tulloch, Research Engineer at Facebook and much more... Opinions expressed in this interview may not represent the views of RE•WORK. As a result some opinions may even go against the views of RE•WORK but are posted in order to encourage debate and well-rounded knowledge sharing, and to allow alternate views to be presented to our community.