A selection of our current research projects can be found below.
Determinantal Point Processes
Chic, shocking, trendy, or casual, Fashion must never be monotonous. As such, diversity plays a central role in fashion. At Zalando we research such topics, focusing on elegant mathematical models named Determinantal Point Processes (DPPs) that capture notions of both relevance and diversity in a principled manner.
Shop The Look With Deep Learning
Have you ever seen a picture on Instagram and thought “oh wow, I need these shoes”? or been inspired by your favourite fashion blogger and looked for similar products (for example on Zalando)? Visual search for fashion, the task of identifying fashion articles in an image and finding them in an online store, has been the subject of an ever growing body of scientific literature over the last few years, and has been offered by many fashion stores.
Personalized Size Recommendation
The aim of this research project is to investigate and advance the state-of-the-art in personalized size recommendation. While Zalando has a lot of data over populations of customers, a key problem with personalization is the shortage of data for individual customers. To alleviate this problem, we are developing models that can leverage information from multiple customer and article channels (e.g., purchase and return streams). On the methodological side, we combine Bayesian theory with deep learning to yield industry’s best personalization experience.
Can we teach machines to read and understand human language? And in the future, can we teach them to perform language AI tasks with little training data, across over 20 different human languages? At Zalando Research, we’re investigating answers to these questions.
Forecasting The Customer’s Preference
We study time series of interactions of articles (fashion items) with customers. Our focus lies on personalized data, meaning, that each time series collects data for a single customer only. Examples include the customer’s purchase history as well as their click chain through the product detail pages in the shop. Our aim is to build a method which takes the past of a time series as input and effectively forecasts future events – the task of a recommender system.
Sample Efficient Reinforcement Learning
Reinforcement learning has seen many successes in the last years, enabling computers to beat human masters at such difficult tasks as Go and Atari games. In addition, reinforcement learning is showing great promise in creating exciting consumer facing technologies such as self driving cars, while also driving behind the scenes efficiency gains with self learning recommendation and pricing systems.
Probablistic Time Series Modelling
This project aims at probabilistic time series models, which is a general problem setting across many modern internet companies. In other words, we want to be capable of getting a predictive distribution for the value of a time series at future time points from our model.
Fashion DNA: Improving Fashion Item Encoding And Retrieval
Unlike a brick-and-mortar department store, online retailers like Zalando must use digital representations of their offerings to entice customers. Traditionally, this takes the form of information stored in a catalog database, which collects categorical properties (tags), numerical descriptors (e.g., prices), and images of the items. Although this information is curated by in-house experts, assignments are often subjective (what makes a dress ‘dark red,’ or ‘leisure’?), and article imagery is highly stylized.
Generative Fashion Design
Modern generative machine learning models allow the sampling of data from a very high-dimensional distribution. A prime example domain is images. Can these methods be used to assist fashion designers and accelerate fashion design? In this project we’re investigating answers to this question.
Swapping Of Fashion On People Images With Generative Modelling
Our work focuses on using state-of-the-art-image modelling techniques in order to flexibly modify the appearance of human fashion images. Our fashion applications of generative modeling require the generation of human body poses conditioned on specific inputs -- e.g. the swapping of clothes that people are wearing, or changing the body pose of a human image while keeping a plausible appearance.
The FEIDEGGER (fashion images and descriptions in German) dataset is a new multi-modal corpus that focuses specifically on the domain of fashion items and their visual descriptions in German. The dataset was created as part of ongoing research at Zalando into text-image multi-modality in the area of fashion.
Generative Models Research
The overarching goal of this project is to advance the field of generative modeling, i.e. to build methods that learn complex probability distributions. In particular, we focus on generative adversarial networks (GANs), a relatively new technique that has shown impressive results in purely data-driven modeling of images.
Fashion-MNIST is a dataset of Zalando’s article images consisting of a training set of 60,000 examples and a test set of 10,000 examples. Each example is a 28×28 grayscale image, associated with a label from 10 classes. Fashion-MNIST is intended to serve as a direct drop-in replacement of the original MNIST dataset for benchmarking machine learning algorithms.