Fashion DNA - a Structural Feature Mapping of Fashion Articles
The universe of fashion articles is a heterogenous set of items with most different individual properties. For this set, a meaningful structure needs to 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, similar items being close by. Fashion DNA derives this structure from a priori available information about fashion products. This comprises product images, textual descriptions, vendor product attributes or combinations thereof. Main building blocks are Deep Neural Networks which process all available information and create for every item a unique coordinate vector with the above mentioned property of encoding similarity. With the help of Fashion DNA articles can be identified (even hypothetical ones that don't yet exist), styles can be described formally, and order can be brought into the chaotic universe of fashion articles.
Roland is the Research Lead at Zalando Research and obtained his Ph.D. at the Technical University of Berlin in Machine Learning and Statistical Signal Processing. Roland was integral to the establishment of Zalando Research and has been with Zalando since 2013. He previously worked as Head of Research for GA Financial Solutions GmbH and conducted the development of asset risk models and quantitative trading strategies.