Programme

Scaling artificial intelligence is one of the biggest challenges for companies in dealing with AI projects. AI disruptively changes companies, their processes and, at its core, impacts employees and customers. From practical experience, trust is a key differentiator to drive AI adoption and scale successfully. The presentation will provide insights from practical applications: the main differences of AI projects that make it difficult to reach production, success factors for scaling AI solutions and the importance of gaining trust therefor. You will learn from real client examples; how trustworthy AI is approached holistically.

Data, analytics & AI are key factors for an efficient and sustainable future at Deutsche Bahn. At DB Netz we want to contribute at the foundation - the infrastructure - by analyzing tunnels, stations, bridges, overhead wires,… That’s why we have kickstarted a new department two years ago with an interdisciplinary team of different experts from DB and Birds on Mars and have already gained a lot of experience by connecting the dots and creating new intelligent architectures in this unique surrounding. In their talk Nina and Florian want to share first insights and talk about a new data product which connects sniffer dogs, data and intelligence to provide creative ways for environmental protection - how? Fly by and find out.

Big Data is getting ready for business, but are businesses ready for BigData/AI? As companies seek to capitalize on big data applications, learn how to leverage vulnerable external data sources while maintaining a secure, controlled digital infrastructure where data can be shared and analyzed.

Klaas Mertens will be demonstrating in this session:

• The power of a neutral infrastructure that ensures security and control for all parties,

• Why data marketplaces can offer a new model for value creation,

• How sandboxes can act as a neutral place where data and algorithms meet,

• Use cases and best practices where Equinix's extended architecture has already been field tested.

At the time of submission, the data portal of the European Union at data.europa.eu comprised 1,396,453 datasets from 36 countries. If a data scientist were to look at each of these datasets for merely one hour, he/she would busy for just short of 700 years. At the same time, data is often considered to be the key raw material of the 21st century, in particular public data that could drive public-sector decisions to the benefit of citizens across the EU.

So who will ever analyze all of Europe’s data? And what about the other national and private-sector data repositories, e.g., data.gov or data.world? From the magnitude of the task, it becomes apparent that this cannot be done without automation—and this is where AI gets to shine! Teaming up with stakeholders both from the public data and the smart cities communities, we have launched a cognitive service that intelligently analyses public datasets. Hosted at analyst-2.ai, this service is working through publicly available datasets, with results now becoming available at everyone’s fingertips, e.g., here.

In this talk, we will cover both underlying technology as well as the considerations from the point of view of the 52 communities that these datasets are about. After an introduction to the public data landscape, our talk will give a look under the hood of how AI-driven autonomous data analysis was applied to the data repositories mentioned above. We will then switch perspectives to address the most pressing requirements and choices that communities have, that may benefit from a more data-driven approach.

This project led to multiple learnings that we will share with the audience, incl. on sensible use of meta-data, anchoring of hypotheses, and efficiently prioritizing and communicating results. We feel that these learning may also be of interest to the business community, since it too now has more data than could ever be analyzed by their analytics teams.

Deutsche Bahn's exciting journey full of experiences, setbacks and out-of-the-box thinking towards a functioning "AI factory" that produces AI use cases and data products in line with the business units with measurable effects while supporting the people in the system to master the increasing complexity.

Both Data Fabric and Data Mesh are currently hot topics in analyst papers, customer discussions and vendor offerings. And both promise impressive value propositions, but are based on quite contrary architectural setups. Even based on concrete company status and requirements the decision for one or the other is difficult. This session will cover definitions, pros and cons and also approaches for combining both architectures. The real value can only be achieved by an evolution based on existing Data Warehouses and Lakes to a tailor-made architecture for your company.

Data Mesh is a promising new concept of how complex corporations should approach data management from the technology as well as the people perspective. But how does a group of corporations, consisting of very different and independent companies, such as Deutsche Bahn group, build up sustainable momentum to fuel this sociotechnical transformation?

Deutsche Bahn follows a lean approach of learning from concrete implementations, so that it can use objective data to guide the transformation and build a storyline for the necessary cultural changes towards Data Mesh. In this talk showcases how a pilot implementation project was used to learn how the approach fits to the company and derive tangible implications for developing Deutsche Bahn’s data management further towards a Data Mesh.

Open-source data ecosystems characterized by data sovereignty, security, interoperability and modularity are the central goal of Gaia-X. Such a dataspace for maritime digital data is being created in the Gaia-X lighthouse project Marispace-X. Gaia-X compatible structures around data acquisition and processing combined with modern open source Big Data technologies enable the implementation of a wide range of use cases: The Internet of Underwater Things use case aims to enable sensor technology to capture a digital twin of the underwater world in real time. Offshore Wind aims to develop data-driven applications and generate trustworthy data spaces for collaboration and efficiency gains of offshore wind farms by collaborative development of tools to map the entire process chain in data collection and analysis.

German waters alone contain over 1.5 million tons of old munitions. As part of the use case Ammunition in the Sea we will develop new smart ways of data analysis and management in cloud, edge, and fog computing to tackle this problem. The goals of the use case Biological Climate Protection are to identify and quantify the potentials of using CO2 storage capacities of oceans plants by establishing novel analysis methods of sensor fusion of optical satellite, 92 remote sensing and underwater acoustic data and through the development of AI-based predictors of optimal macrophyte settlement areas.

A very heterogeneous data landscape, computationally intensive data analysis and reduced data transmission capabilities underwater require new directions in maritime big data processing and analysis of sensor data via edge, fog and cloud computing. The integration with Gaia-X Federation Services ensures compliance with the open Gaia-X standards at the infrastructure, data platform and services levels. In this way, the concepts and implementations developed in Marispace-X can also be transferred to other domains.

Railbound mobility and logistics are taking on a central role in Germany due to increasing environmental awareness. The resulting increase in traffic volume is a key challenge that Deutsche Bahn is addressing, among other things, through the use of AI. The presentation shows how AI can help improve processes before, during and after train travel. Data and AI form the basis for networking processes from dispatching to maintenance. Thereby, the complete "zoo" of machine learning - ranging from Supervised to Reinforcement Learning - is used with the aim of improving quality, flexibility and capacity. An insight into our strategic approach forms the outlook of the presentation.

Artificial intelligence (AI) has the potential to fundamentally change our society and our working world. As a result of the desired autonomy and decision-making power of computer-controlled systems, the explainability of AI represents one of the central challenges for the broader use of the new technology. The more powerful and efficient algorithms become, the less comprehensible it usually gets for users to understand how they make their decisions. To better understand the current state of the industry, PwC conducted a survey on the adoption rate of Explainable AI across major financial services institutions. While there is widespread agreement on the basic requirement of explainability, the design of an appropriate explanation is rarely precisely defined. To mitigate this, we created a holistic framework to assess an organization’s Explainable AI readiness and assist in operationalizing Explainable AI. Combining organizational, regulatory and technical dimensions in a unique way, the framework allows organizations to assess their adoption of Explainable AI from a management perspective while also providing the means to algorithmically evaluate the explainability of currently used AI models. In the presentation we will further explore those dimensions of explainability and take a deep dive into topics such as performance, comprehensibility, granularity and faithfulness.

After the financial crisis, Europe’s leading banks and rating agencies worked with digital business transformation company Publicis Sapient to build European DataWarehouse (EDW), a market-owned central data repository endorsed by the European Central Bank (ecb). In 2012, EDW was founded as part of ecb’s ABS loan level initiative to restore confidence in the securitisation market. The aim was to provide all market participants with access to standardized, timely information on ABS deals and private loan portfolios and their performance.

All securitisations at individual loan level as well as the relevant documentation have been recorded and extensively quality-checked for the last ten years. The data is openly accessible and allows market participants to make informed risk assessments and decisions at any time based on detailed, up-to-date and asset class-specific information.

At the core of the repository architecture, EDW and Publicis Sapient are leveraging Microsoft’s Azure cloud big data platform with continuous delivery powered by DevOps. This repository platform supports collection, real time 82 processing/validation and storage of the securitisation-related billions of loan level data records along with assessment of data completeness and quality.

In 2021, EDW was registered as one of the first Securitisation Repositories (SRs) by the European Securities and Markets Authority (ESMA), and in 2022 as one of the first UK SRs by the Financial Conduct Authority (FCA). Developed, owned, and operated by the market, EDW with its extensive data treasure trove has become a true European data champion whose importance now extends far beyond the securitisation market.

Dr. Thun, European DataWarehouse CEO, and Dr. Leibrock, Managing Partner at Publicis Sapient, will explain the cloud big data platform, discuss its extensive potential for risk mitigation and demonstrate the scope of possibilities that this treasure trove of data opens up for market participants.

At EnBW, we believe in the power of data and AI, which we use in more than 60 use cases to drive the Energiewende. As we expand our activities, a transition from focusing exclusively on use cases to a comprehensive approach that empowers the entire organization is required. In addition to technical measures such as data quality, data access, and citizen data tools, we are placing an emphasis on the cultural change that comes with the transition to a data-driven organization. A key success factor for us is training our employees in both technical and business-oriented roles. Our vision for data & AI includes the phrase "Handling data comes completely naturally to all of our colleagues."

In our presentation, we will outline how we have organized our company-wide educational activities to make working with data natural. However, the focus of our presentation will be on our custom-designed Data Professional Development Program, (DP)2, which aims to further enhance the skills of our best colleagues in data-related roles. (DP)2 is a year-long program that brings together 20 data engineers and analysts from across the company and is
based on four pillars: First, we organize group training sessions (e.g., in design thinking) for the entire cohort to give them the opportunity to dive into a new topic together. In addition, each participant is given a budget to book 4 trainings that fit their individual learning path. Second, because participants do not work with each other on a daily basis (or even know each other), we create a community where they can connect and learn from each other. Third,
at this skill level, understanding best practices is a primary interest of particip ants. Therefore, we bring them together in joint projects that allow them to learn about different ways of working, techniques, and team cultures. The fourth pillar is to bring external insights into the program to learn from other companies.

Is it possible to predict and prevent residential burglaries using AI ? In this presentation we will show how we built an artificial intelligence system for predicting residential burglaries in Berlin together with the Berlin police. The solution “Krimpro” is in productive use in Berlin since 2017 and supports the detection and prevention of the residential burglary in the capital. Learn what practical benefits the investigating authorities derive from the system and what the important success factors for doing an AI project in the public sector are.

We are at a unique point in medical history where a wide variety of modern analytical methods and digital technologies are coming together. Wearables, the smartphone or clever apps for recording one's own vital data provide a large amount of relevant health data. These data volumes, which are summarized under the term Smart/Big Data, hold great potential. A human being, however, can hardly do anything with this volume of data. On the other hand, for computers, this digital information is a valuable resource that - once standardized - can be analyzed.

AI systems are already analyzing gigantic amounts of data - a few of which are being used in medicine. Whether in diagnosis, early detection, therapy, as decision support or in research - AI in medicine supports healthcare by taking on routine tasks, as well as providing the knowledge to support those tasks.

The CovidDataNet.NRW project shows how "smart data" and artificial intelligence are being used in an attempt to better understand the course of diseases and to intervene in an individualized way. After the onset of the Corona pandemic, it quickly became clear that many of the severe Covid-19 infections were viral-induced sepsis. The inflammatory response, which is the third leading cause of death in Germany, has not been researched comprehensively enough using commercial methods to characterize any patterns or develop clinical testing procedures. Therefore, the CovidDataNet.NRW project brings together multidisciplinary institutions (intensive care/anesthesia, immunology, protein research, Big Data management) throughout North Rhine-Westphalia. The partners in the consortium are attempting to decipher the complex system of sepsis mathematically using pattern analyses with the aid of state-of-the-art IT technologies, a Big Data platform and artificial intelligence. The hope is utilize big data to optimize diagnosis and subsequent treatment of sepsis.

Long before Big Data was a thing, governments, business and NGOs relied on public opinion polls to effectively deal with both political and commercial questions of our time. Raider vs. Twix? D-Mark vs. Euro? Lock-down vs. 3G+? Surveys provide evidence to drive critical decisions.

The statistical methods employed by political/social sciences both effectively deal with biased data and extract meaningful information even from sparse datasets. Could AI reliably come up with better projections than a trained human statistician? Not in the general case! But an AI can easily process data faster, more comprehensively and with a more detailed step-by-step account than any human.

In this talk, we relate how Inspirient and Kantar Public Germany (formerly TNS Infratest) joined forces in an effort to emulate the reasoning steps of human statisticians using AI building blocks, effectively building a “junior AI statistician” with a deep enough understanding of the subject matter and relevant methods to autonomously analyze any real-world survey dataset. This endeavour is far from complete, but our system is up and running and helping Kantar Public’s expert statisticians and researchers to get meaningful results faster than ever before.

Our presentation briefly introduces existing analytical processes at Kantar Public and Inspirient’s architecture for autonomous analytical reasoning. We then dive deep on how Kantar Public’s expert statisticians and Inspirient’s AI engineers jointly modelled and implemented the necessary reasoning steps, so that the chain of classifiers can derive analytical results that can be trusted, just as if a human had done the calculation. We then share our observations and learnings from initial business use: Through the system’s comprehensiveness, the AI-generated results become a second opinion on what key insights from a survey may be, very much complementing the experience-driven approach of the seasoned statistician.

Trees are indispensable in mitigating the effects of climate change and helping cities’ climate change adaptation through shade, transpiration, and support of biodiversity. However, city trees are stressed and dying due to longlasting droughts and decreasing rainfalls. This talk starts with a general overview of opportunities for AI in climate change mitigation and adaptation and presents the publicly funded project “Quantified Trees” as an example application. We show how we create scalable AI tools for the effective watering and care of city trees by using machine learning to bridge the gap between local IoT sensors and physics-informed soil water balance models. Quantified Trees connects citizens, public authorities, and experts on botany and AI to ensure that AI is not simply used for its own sake but to create sustainable solutions.

The AI Act is coming out soon. New use cases are categorized as high risk and must meet technical requirements. Alternative standards and regulations are under development. A systematic risk assessment will be required on high-risk machine learning products.

Florens Greßner is going to talk about neurocat's learnings from five years of experience in implementing testing strategies for safety and security critical domains such as automotive, biometrics and manufacturing.

Energy efficiency gets nowadays more and more important within the utility sector. The presentation is focused on a use case within water management where typically several compressor systems are used in parallel to aerate clarifiers.

The potential of AI assisted load sharing is an optimization algorithm that can significantly decrease the power consumption of the air supply compared to classic approaches. The AI uses a digital twin of the system to simulate the performance and uses evolutionary algorithms to find the control parameters with the lowest power consumption.

Federated learning is a rising AI technology. However, usually Federated Learning is associated with horizontal federated learning - means same features but different data. Vertical federated learning also known as split learning brings different features with the same data together. Katulu applied this approach to whole value chains within the industry, so that all parties within a value chain can learn and benefit from each other without learning about each other / sharing thei r data. In this talk I would like to outline the concepts behind split learning and how it can be transferred to industrial processes across multiple companies.

In autonomous driving, sensor data is used, among other things, to train AI models. Up until recently, in order to make the data usable, all of it had to be collected centrally in the cloud. This places high demands on infrastructure and user privacy and results in unnecessary data being collected, for example, driving on uneventful motorway journeys. Federated Learning and Fleet Learning are new approaches that either prevent data collection altogether or only collect relevant data from the vehicle fleet. AI models are used directly on the vehicle to assess data according to relevance, which in many cases makes data transmission superfluous and leads to a more valuable data set on the cloud side. In certain cases, there is even the possibility of performing the training directly on the vehicle. In this session, Daniel Beutel will introduce the concept of fleet learning and demonstrate how fleet learning can be applied with currently available frameworks.