Deep Learning in Retail & Advertising Summit schedule

Using deep learning to estimate Customer Life Time Value (CLTV) in e-commerce

CLTV prediction is an important problem in e-commerce. An accurate estimate of CLTV allows retailers to correctly allocate marketing spend, identify and nurture high value customers, minimise exposure to unprofitable customers and attribute value to indirect marketing such as content production. We describe how ASOS combines automatic feature learning through deep neural models with hand-crafted features to produce CLTV estimates that outperform either paradigm used in isolation.

Ben Chamberlain is a senior data scientist at ASOS.com where he leads the Customer Understanding team. He holds a Royal Commission for the Exhibition of 1851 Industrial Fellowship, which funds his PhD studies in statistical machine learning at Imperial College London. Ben has previously worked as a data scientist in the social media, defence and security industries. He is a graduate of the University of Oxford.

Deep Learning in Micro-Moment of Shopping

Today every single retailer is harnessing the power of Big Data and Machine Learning to understand their customers and deliver the best shopping experience. These experiences can be categorised into a broad range in a customer's lifecycle. In my talk, I will discuss one of the category; the Micro-Moment of the Shopping Experience. I will elaborate on how a retailer can leverage computer vision and natural language processing to help their consumers make smart decisions in their mico-moments of the shopping experience.

Kumar Ujjwal is a Sr. Product Manager for Big Data and Machine Learning products at Kohl's Department Store. His current work is focused on using computer vision and natural language processing for customer engagement in the retail industry. Previously, he Co-founded two companies incorporating data science and Machine Learning in hospitality and transportation industry. Kumar is a big believer in continuous learning, and therefore he has always involved himself in online learning and open source projects.

DeepAR: Probabilistic Forecasting with Autoregressive Recurrent Networks

A key enabler for optimizing business processes is accurately estimating the probability distribution of a time series’ future given its past. Such probabilistic forecasts are crucial e.g. for reducing excess inventory in supply chains. In this talk, I will present some of the forecasting challenges at Amazon and then introduce DeepAR, a novel methodology for producing accurate probabilistic forecasts, based on training an auto-regressive recurrent network model on a large number of related time series. We show through extensive empirical evaluation on several real-word forecasting data sets that our methodology is more accurate than competing state-of-the-art models, without requiring any manual feature engineering.

Jan Gasthaus is a machine learning scientist in Amazon’s Core Machine Learning team, working mainly on time series forecasting and large-scale probabilistic machine learning. He is passionate about developing novel machine learning solutions for addressing challenging business problems with scalable machine learning systems, all the way from scientific ideation to productization. Prior to joining Amazon, Jan obtained a BS in Cognitive Science from the University of Osnabrueck, a MS in Intelligent Systems from UCL, and pursued a PhD at the Gatsby Unit, UCL, focusing on Nonparametric Bayesian methods for sequence data.

Machine (Deep) Learning and Its Applications to Hotel Search Problems

Hotels metasearch engines such as trivago serve millions of queries daily. One of the challenges facing such search engines is to predict the intention of users queries. For instance, when a user types “Hotels in Munich” we might want to recommend “Oktoberfest” as an additional search keyword. Recently, machine/deep learning learning methods have shown a great performance in several domains including natural language processing and computer vision. In this talk, I would like to present a framework based on word embedding and it aims to compute reasonable recommendations for our users. Next, I would like to talk about future work, where we will apply deep learning approaches to classify hotels to provide better search facilities.

Rami Al-Salman. Al-Salman received a B.S. and M.S. in computer engineering from Jordan University of Science and University (JUST) in 2008 and 2011, respectively and a PhD in computer science from Bremen University, Bremen, Germany in 2014. He is a data scientist and machine (deep) learning engineer at trivago, where he has been since 2015. In this role, he is developing the next generation of recommendations engines that involve heavily AI and machine learning techniques such as Deep Learning, ANN, NB, SVM etc that will allow to capture the intention of the users queries. He worked as researcher and machine learning engineer in 2014 for a space company called OHB in Bremen, Germany. His research interests are in artificial intelligence, machine learning, and natural language processing. He published several well cited articles.

ML for Online Supermarkets - From Convenient to Smart Shopping

Picnic is the world’s fastest growing online supermarket that makes grocery shopping simple, fun, and affordable for everyone. We will show you how we transformed an already convenient shopping experience into a delightful ultra-fast shopping blitz-stop. In this talk we provide a view behind the scenes of our deep-learning based behavioral analytics and prediction engine. We will talk you through our ups-and-downs of product, category and promotional recommendations of FMCGs and do not shy away from demoing also the failures. Now we are able to predict with >95% likelihood the top 12 articles of the next order of each of our customers.

Daniel Gebler is CTO of Picnic, the world’s fastest growing online supermarket that makes grocery shopping simple, fun, and affordable for everyone. Previously, he was Director R&D of Fredhopper, responsible for the product and technology roadmap, and led engineering teams located in Amsterdam and Sofia. Daniel holds a PhD in Computer Science and an MBA. He is also a dad of two young kids, used be a passionate skater and nowadays an enthusiastic climber!

Deep learning for retail warehouse operations

Recent advances in deep learning have enabled research and industry to master many challenges in computer vision and natural language processing that were out of reach until just a few years ago. Yet these challenges represent only the tip of the iceberg of what is possible. In this talk, I will demonstrate how we have used deep neural networks in to solve a special case of the travelling salesman problem and in turn steer operations at Zalando’s fashion warehouses. Further, I will present a few tips and tricks on getting GPU enabled neural networks running with minimal technical overhead.

Calvin Seward is a Resarch Scientist for Zalando's Research Lab, working mainly on unsupervised and semi-supervised computer vision problems including classification, localization and semantic segmentation. The main line of attack for these problems is deep neural networks written in the Tensorflow framework and trained on GPU clusters. At the same time, he is involved in applying lessons from GPU-driven HPC computing and cutting edge Machine Learning to other fields of the Zalando universe. Past projects have included sizing solutions for online retail, warehouse management and recommendation systems.

Pau Carré is a data science and software engineer at Gilt. Pau has 10 years of experience encompassing software security, IT management, microwave networks profiling, quality engineering and currently deep learning in fashion industry. He holds a Bachelors and Masters degree in Computer Sciences from the Universitat Politècnica de Catalunya.

At Gilt, we are reshaping the fashion industry by leveraging the power of GPU accelerated deep neural networks hosted on Amazon Web Services. Our deep neural network classifier uses product images and text description to predict product facets, such as silhouette type for dresses or heel type for shoes. Additionally, our deep learning similarity system detects visually similar products, extracting features from product images and storing them in a dense vector database for fast retrieval. Through our faceting and visual similarity detection, we’re able to improve product recommendations and product filtering for our customers. Come learn about the architectures and theory powering our systems.

Stuart has recently joined the John Lewis Partnership having set up an Artificial Intelligence capability in Thames Water and has experience in delivering AI tools in both the utilities and retail industry. Stuart is looking to quickly deliver innovative tools and develop a new way of working to both Waitrose and John Lewis in an extremely competitive market.

Henry is a doctoral candidate at the University of Warwick in the UK with a research focus on theoretical cognitive science and robotics. His research focuses on the ways in which low-level dynamic and computational control of motor action can affect human-human and human-robot interactions. He is also part of the ERC-funded "Sense of Commitments Project" which takes place at The University of Warwick in the UK, The Italian Institute of Technology in Genoa, and CEU in Budapest. For this project, Henry carries out research into how human-like commitments can be engineered into social-robotics platforms to engender beneficial human-robot interactions in medical, commercial, and social spheres.

Deep Learning in Micro-Moment of Shopping

Today every single retailer is harnessing the power of Big Data and Machine Learning to understand their customers and deliver the best shopping experience. These experiences can be categorised into a broad range in a customer's lifecycle. In my talk, I will discuss one of the category; the Micro-Moment of the Shopping Experience. I will elaborate on how a retailer can leverage computer vision and natural language processing to help their consumers make smart decisions in their mico-moments of the shopping experience.

Kumar Ujjwal is a Sr. Product Manager for Big Data and Machine Learning products at Kohl's Department Store. His current work is focused on using computer vision and natural language processing for customer engagement in the retail industry. Previously, he Co-founded two companies incorporating data science and Machine Learning in hospitality and transportation industry. Kumar is a big believer in continuous learning, and therefore he has always involved himself in online learning and open source projects.

Hussein is a founding partner of Hoxton Ventures, a London-based early stage European venture capital firm. He currently represents Hoxton on the boards of bd4travel, Darktrace and TourRadar. Hussein is an angel investor in Bea's of Bloomsbury, GoCardless, Halo Neuroscience, Popxo, Signpost and Yieldify. He is an advisor to Tarmin, and was previously an advisor to the chairman of Eros International (NYSE:EROS). He has occasionally assisted the UK government with technology policy. He regularly appears on BBC, Bloomberg, CNN and Sky News. Formerly, Hussein was an associate at Accel Partners, where he focused on consumer internet, financial technology and software investments. He helped invest in OpenGamma, Playfish (acquired by EA) and Dapper (acquired by Yahoo). Hussein joined Accel from Microsoft Corporation where he was a platform planner, product manager and senior manager. He also helped put together Safe-View (acquired by L-3), and worked at Radiance Technologies (acquired by Comcast) and Studio Verso (acquired by KPMG). Hussein holds an MBA from London Business School, and did his undergraduate studies in Symbolic Systems at Stanford University.

Maintaining Generic ConvNet Representation when Switching Data Type

Volumental is a Swedish computer vision company bringing 3D scanning and fitting solutions to footwear retailers around the world. To provide consumers with the most intelligent recommendations and engaging shopping assistance, one part of our product is about accurately measuring the consumers' feet. We use deep learning at various levels of our pipeline to represent, segment and classify visual and shape data. This talk will focus on how we have made use of powerful convolutional nets on data very different from what they were originally trained for, significanly reducing the annotation effort needed to reach high accuracy on new types of visual data.

Miroslav aqcuired his MSc in Optics & Photonics from Imperial College in 2008 and has since then worked with R&D in computer vision and machine learning in the industry and academia. He co-founded Volumental during his PhD in computer vision at KTH and is currently leading the company's deep learning efforts in building the world's best and most consumer centric sales assistant.

Deep Learning for Fashion Attributes

The fashion industry is a visual world. Millions of images are displayed everyday by fashion commerce sites to serve consumers the latest trends and products. However, automatically categorizing and searching through large collections of images according to fine-grained attributes remains a challenge. In this talk we present our research on deep learning techniques to automatically identify fine-grained attributes in both images and text in the presence of incomplete and noisy data. We focus on attributes such as necklines, skirt and sleeve shapes, patterns, textures, colors and occasions. This task is especially useful to online stores who might want to automatically organize and mine visual items according to their attributes without human input. It is also useful for end users who wish to find specific items when there is no text available describing the target image. We compare the results of the deep learning approach with classical techniques such as Latent Dirichlet Allocation and Canonical Correlation Analysis. Our results show that it is possible to design algorithms that automatically “translate” visual concepts into text and vice-versa.

Susana Zoghbi received a PhD in Computer Science in December 2016 from the KU Leuven university in Belgium. Her research interests lie at the intersection of computer vision and natural language processing, and include deep learning, topic modelling and probabilistic graphical models. During her PhD, she developed latent-variable models for understanding language and images in social media and e-commerce data. She holds two Master’s degrees, one in Mechanical Engineering, where her research focused on human-robot interaction technologies, and one in Mathematical Physics, where she focused on gravitational fluctuations in Domain Wall Spacetimes. In 2014, she was awarded a Google Anita Borg Scholarship. More information can be found at: https://people.cs.kuleuven.be/~susana.zoghbi/

Fashion Product Retrieval with Real World Images

Online retail stores have vast catalogs with hundreds of thousands or even millions of products. Searching for the perfect product in this space with basic text search would already be a daunting task, but it gets worse: more often than not, product descriptions are inadequate or they do not exist at all. If you compound different languages and vocabulary choices between different retailers and geographical regions it only gets more difficult. On the other hand, images are a universal language that, with current deep learning techniques, can be leveraged to search for similar products without the limitations of text. They have, however, their own set of difficulties: catalog images are taken by professional photographers in ideal conditions, often with a white background, while "query pictures" are much more diverse and can have many undesirable characteristics such as bad illumination, motion blur, complex backgrounds or unusual viewpoints. To bridge this gap, we can use a type of model called Siamese networks that, using pairs of corresponding shop and consumer pictures, learn a common embedding for both styles of image.

Arnau Ramisa received the MSc degree in computer science (computer vision) from the Autonomous University of Barcelona (UAB) in 2006, and in 2009 completed a PhD at the Artificial Intelligence Research Institute (IIIA-CSIC) and the UAB. Between 2009 and 2011, he was a postdoctoral fellow at the LEAR team in INRIA Grenoble / Rhone-Alpes, and between 2011 and 2015 a research fellow at the Institut de Robòtica i Informàtica Industrial in Barcelona (IRI). Since 2015 he is working as a computer vision researcher at Wide Eyes Tech. His research interests include object classification and detection, image retrieval, robot vision and natural language processing.