Machine Intelligence Summit schedule

Characterising extrasolar planets with deep learning

The discovery of extrasolar planets - i.e. planets orbiting other stars - has fundamentally transformed our understanding of planets, solar systems and our place in the galaxy. With over 3500 systems discovered to date, characterising these alien worlds is rapidly becoming a big data problem. Future surveys with ground and space based telescopes will simply provide too much data to be analysed in more classical ways. Here I will present how deep learning can be used to rapidly characterise the chemistry and prevalent weather patterns of these extrasolar planets and put our own solar system in the grander galactic context.

Dr. Ingo Waldmann is a senior research scientist at the University College London. He obtained his PhD in astrophysics in 2012 working on blind source-separation algorithms applied to observations of extrasolar planets with the Hubble and Spitzer space-telescopes. He has since specialised in the modelling of non-linear Bayesian inverse problems and deep learning applied to atmospheric physics of extrasolar planets and solar system objects. He is the data analysis lead of the European Space Agency ARIEL mission and the UK-led Twinkle space-mission.

Exploring the Latent Visual Space Between Adjectives with Generative Adversarial Networks

Generative Adversarial Networks (GANs) have been applied for multiple cases, such as, generating images and image completion. One interesting feature of GANs is the exploration in the Latent Space, where new elements can appear caused by the interpolation between two seed elements. With this in mind, we are interested in exploring the latent space in terms of Adjective-Noun Pairs (ANP) able to capture subjectivity in visual content such as "cloudy sky" vs. "pretty sky". Although it is challenging for humans to find a smooth transition between two ANPs (similar to color gradient or color progression), the presented GANs are capable of generating such a gradient in the adjective domain and find new ANPs that lies in this (subjective) transition. As result, GANs offer a more quantified interpretation for this subjective progression and an explainability of the underlying latent space.

Dr. Damian Borth is the Director of the Deep Learning Competence Center at the German Research Center for Artificial Intelligence (DFKI) in Kaiserslautern, the Principle Investigator of the NVIDIA AI Lab at the DFKI, and founding co-director of Sociovestix Labs, a social enterprise in the area of financial data science. Damians research focuses on large-scale multimedia opinion mining applying machine learning and in particular deep learning to mine insights (trends, sentiment) from online media streams.

Augmenting Search Quality Ratings with Logs Data

When working on a search engine, we need to measure how our improvements contribute to user satisfaction. While human ratings are often used as a proxy for that, they are costly and do not always correlate well with the user-reported satisfaction. In this talk I present a machine learning model that allows us to leverage log data to build a more accurate rater-based quality metric. Combining log data—which is easily collectible at scale—with a limited-sized raters’ labels allows us to get an estimation of user satisfaction for previously unseen document layouts at no additional cost.

Aleksandr is based at Google Research Europe, Zurich Switzerland, where he is currently working on the problems related to search-powered digital assistants. Aleksandr has a PhD from University of Amsterdam, where his research was focused around modeling and understanding user behavior on complex search engine result pages. He has a number of publications in leading information retrieval conferences and journals, and co-authored a book and multiple tutorials on click models for web search. Prior to Google, Aleksandr worked on search-related problems in Applied Research group at Yandex.

Learning Semantic Environment Perception for Cognitive Robots

Robots need to perceive their environment to act in a goal-directed way. While mapping the environment geometry is a necessary prerequisite for many mobile robot applications, understanding the semantics of the environment will enable novel applications, which require more advanced cognitive abilities. In the talk, I will report on methods that we developed for learning tasks like the categorization of surfaces, the detection, recognition, and pose estimation of objects, and the transfer of manipulation skills to novel objects. By combining dense geometric modelling – which is based on registration of measurements and graph optimization – and semantic categorization – which is based on random forests, deep learning, and transfer learning – 3D semantic maps of the environment are built. Our team demonstrated the utility of semantic environment perception with cognitive robots in multiple challenging application domains, including domestic service, space exploration, search and rescue, and bin picking.

Prof. Dr. Sven Behnke is a full professor for Computer Science at University of Bonn, Germany, where he heads the Autonomous Intelligent Systems group. He has been investigating deep learning since 1997. In 1998, he proposed the Neural Abstraction Pyramid, hierarchical recurrent convolutional neural networks for image interpretation. He developed unsupervised methods for layer-by-layer learning of increasingly abstract image representations. The architecture was also trained in a supervised way to iteratively solve computer vision tasks, such as superresolution, image denoising, and face localization. In recent years, his deep learning research focused on learning object-class segmentation of images and semantic RGB-D perception.

Deep Learning Workload and its Impact on Hardware Requirements Throughout the AI Product Lifecycle

The goal of this talk is to review infrastructure challenges for supporting deep learning workloads, and the impact they have on the research and AI product development cycles. In particular, the talk will touch on three main areas: • The deep learning workflow and the hardware requirements for each stage of the product development process • Hardware requirements associated with deep learning model development • Current trends in deep learning and their impact on the future hardware requirements

Adam is an applied research scientist specializing in Machine Learning, with backgrounds in Deep Learning / System Architecture. He is currently a Deep Learning Solution Architect at NVIDIA where his primary responsibility is to support a wide range of customers in delivery of their deep learning solutions. In his previous role with Capgemini he was responsible for building up the U.K. government’s Machine Learning capabilities. He also worked in Jaguar Land Rover Research Centre and was responsible for a variety of internal and external projects, contributing to the ‘Self Learning Car’ portfolio, specifically.

Using Deep Learning to Support Customer Operations

An insight into how Monzo is using deep learning to empower all the customer support, across the bank's various departments.

Utilizing Sentence Embeddings to Extract High Coverage Key-Phrases

Extracting key-phrases is a Natural Language Processing task which has been studied for decades. What makes the task difficult is its intrinsic subjectivity and the fact that we cannot rely on labelled data. Traditionally, extracting key-phrases is tackled by graph based algorithms. This presentation aims to introduce an alternative technique, which leverages on recent research. Instead of using graph metrics, this novel technique relies on sentence embeddings to capture the main points of a document. A robust unsupervised technique, using compositional n-Gram features, is applied to form the embeddings. Finally, this method is incorporated into one of our core products with much success.

Sotirios Fokeas is a Machine Learning specialist working for Swisscom. He is located at the Innovation Park in Lausanne, where he remains at close contact with EPFL's research teams. Sotirios obtained his master's degree in Computer Science from EPFL with a focus on Machine Learning. Prior to Swisscom, he worked as a data scientist in the banking industry, where he developed a new approach for countering Money Laundering. Sotirios research is now focused on unsupervised techniques for analysing and extracting information from large volumes of textual data.

Recording The Visual Mind: Understanding Aesthetics with Deep Learning

With the rise of mobile cameras, the process of capturing good photos has been democratized - and this overload of content has created a challenge in search. One of the important aspects of photography is that every image communicates with a different audience in different form. This talk will address how we use computer vision techniques at EyeEm measure visual aesthetics in photography -and beyond that- personalize the image search experience to find the photos you personally find beautiful.

Appu is the Head, Research & Development at EyeEm. His first company, sight.io, was acquired by EyeEm in 2014 and also held post-doctoral positions at EPFL working alongside Prof. Sabine Süsstrunk and Prof. Pascal Fua. In 2009, Appu obtained his Ph.D. from IIT Bombay, where he was awarded best thesis from Computer Science Dept. He was also selected as one of the most promising 20 entrepreneurs of Switzerland in 2013. His research has appeared in top computer vision journals and conferences such as TPAMI, CVPR, and ACM Multimedia etc.

Deep Learning through Space and Time for Document Image Analysis

Document Analysis and Text processing are essential in our daily life for gather and access knowledge. Starting with the successful adaption of Long Short-Term Memories to Handwriting Recognition, very recent deep learning trends have revolutionized the Document Analysis field in all areas. This presentation gives an overview of the most successful methods in various areas and presents a framework for detection and segmentation of textual information from text-documents, natural, and born-digital (computer generated) images. Finally, the practicability is demonstrated by showing the application of such methods for business form processing in the health-care domain.

Marcus Liwicki, head of the MindGarage, is an apl.-professor in the University of Kaiserslautern and a senior assistant in the University of Fribourg. His research interests include machine learning, pattern recognition, artificial intelligence, human computer interaction, digital humanities, knowledge management, ubiquitous intuitive input devices, document analysis, and graph matching. In 2015, at the young age of 32, he received the ICDAR young investigator award, a bi-annual award acknowledging outstanding achievements of in pattern recognition for researchers up to the age of 40. He has more than 200 publications, including more than 20 journal papers (h-index 25).

Urban Modelling with Big Data and Machine Learning

Machine Learning and Big Data together offer a universal way of looking at the world phenomena, which is radically different than the classical expert based disciplinary research. This new approach of computational modelling has inverted the classical notion of expertise from “having the answers to the known questions” to “learning to ask good questions”, where the answers can always be found with an appropriate level of modelling skills. In this regard, I will show the results of some of our ongoing data driven projects such as urban morphology, real estate market, urban air pollution and urban water flow modelling.

Previously trained and practiced as systems engineer, currently Vahid Moosavi is a senior researcher at the chair for Computer Aided Architectural Design (CAAD), ETH Zurich. In his PhD he was focused on theories of computational urban modelling and issues of “representation” and “idealisation”. Parallel to research and teaching “Data Driven Modelling” to graduate architectural design students at ETH, he has been conducting several applied machine-learning projects such as urban traffic dynamics, urban design, air pollution modelling, networked economy and systemic risk, real estate analysis and recently on scalable emulations of urban water flow.

Characterising extrasolar planets with deep learning

The discovery of extrasolar planets - i.e. planets orbiting other stars - has fundamentally transformed our understanding of planets, solar systems and our place in the galaxy. With over 3500 systems discovered to date, characterising these alien worlds is rapidly becoming a big data problem. Future surveys with ground and space based telescopes will simply provide too much data to be analysed in more classical ways. Here I will present how deep learning can be used to rapidly characterise the chemistry and prevalent weather patterns of these extrasolar planets and put our own solar system in the grander galactic context.

Dr. Ingo Waldmann is a senior research scientist at the University College London. He obtained his PhD in astrophysics in 2012 working on blind source-separation algorithms applied to observations of extrasolar planets with the Hubble and Spitzer space-telescopes. He has since specialised in the modelling of non-linear Bayesian inverse problems and deep learning applied to atmospheric physics of extrasolar planets and solar system objects. He is the data analysis lead of the European Space Agency ARIEL mission and the UK-led Twinkle space-mission.

The Future of AI: Active Learning

The real potential of Artificial Intelligence, is making everyone’s home and work environments smart. Software for work, home and leisure will know greater intelligence, changing itself to your specific person and purpose. Think of medical software learning from doctors, steel inspection software learning from operators and home robots learning from you. The only way to democratize AI is to enable anyone to teach an AI to perform a task. In this presentation, Tijmen Blankevoort, CTO of Scyfer, explains this philosophy, how we should change focus in AI to reach this goal, and ramifications for our future.

Tijmen Blankevoort is co-founder and CTO of Scyfer, a spin-off company of the University of Amsterdam specialized in bringing Artificial Intelligence to business. He advises multinationals and other companies in the Netherlands on Machine Learning, Artificial Intelligence and its practical applications Tijmen can often be found on stage presenting about the newest developments in AI in an enthusiastic and engaging fashion.

Machine Intelligence for Matching People and Jobs

For many people, their job is no longer just a source of income: it is an avenue for personal growth and accomplishments. This puts additional pressure both on the job seeker and on hiring managers or HR professionals. In this talk I will briefly discuss how Machine Learning and Deep Learning are revolutionizing the tooling for matching people and jobs. On the document understanding side, I will show how LSTM neural networks are improving the quality and robustness of Textkernel’s resumes and job ads parsing product. In addition, I will present our recent efforts in building a language-independent resume parsing system akin to Google’s universal machine translation system. I will also discuss how Siamese architectures can be used to learn similarities between job titles (e.g. “java developer” vs “java programmer”). On the searching and matching side, I will describe how relevance can be improved and customized via Learning to Rank.

Mihai Rotaru is the Head of R&D at Textkernel where he is responsible for the research agenda and for coordinating the joint research efforts with the parent company, CareerBuilder. In this role he is guiding 3 teams: the Document Understanding team which builds the resume and job parsing product, the Search R&D team which is improving the Search & Match product and the Ontology team which builds the HR domain knowledge graph. Originally from Romania, he joined Textkernel in 2008 after obtaining a PhD degree in Computer Science at University of Pittsburgh, USA. Among others, he is interested in Machine Learning, NLP, Deep Learning and how these new technologies can be applied in industry.

Julius Ruessmann works as an Investment Analyst with Earlybird Venture Capital in Munich. At Earlybird, Julius focuses on the Mobility Technology Sector, Industrial Technology, Big data and Analytics. Julius has gained experience in the Investment Banking (Mergers & Acquisition) industry as well as in the Private Equity Industry. Through his studies at EBS University and FGV São Paulo he focused on Finance, Strategy and Management and graduated with Bachelor degree.

Carlos Eduardo Espinal is a Partner at Seedcamp, Europe’s first seed fund, identifying and investing early in world-class founders building valuable, global businesses using disruptive technology. Seedcamp has invested in over 240 companies since launch in 2007 and Carlos is an expert in helping startups set fundraising milestones and find product-market-fit.

Prior to Seedcamp, Carlos was an Associate at Doughty Hanson Technology Ventures and an engineer for the Advanced Communications Technologies group of The New York Stock Exchange (SIAC).​ Carlos​​ is an active podcaster and hosts the applauded Seedcamp podcast 'This Much I Know' as well as a published author, with his debut book 'The Fundraising Fieldguide' published in 2015.

From 1-click buying to 0-click fulfillment

First, we will demonstrate how our fleet of real-time connected vehicles, smart planning algorithms, precise monitoring tools and predictive distribution models create an unprecedented distribution logistic system for groceries. Then we provide a look behind the scenes of our smart planning algorithm that broke world records of the famous traveling salesman problem (TSP). Finally, we deliver insights into the architecture and design of our distribution eco-system (planning, monitoring, execution, control components) and explain how we made them ready for autonomous distribution and self-driving vehicles.

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 to be a passionate skater and nowadays an enthusiastic climber!

Matthew worked on trading floors at Lloyds and BarCap before seeing the light. Following an MBA at SDA Bocconi he founded a sunglasses business and was an angel-operator in a couple of science-led startups. At Forward Partners he invests in eCommerce, marketplace and Applied AI companies as well as adding value across the later stage portfolio.