machine learning

Institutions

• UHH

Prof. Dr. Louise Amoore, Durham University, Durham, UK

Institutions

• UHH

Current AI regulation in the EU and globally focus on trustworthiness and accountability, as seen in the AI Act and AI Liability instruments. Yet, they overlook a critical aspect: environmental sustainability. This talk addresses this gap by examining the ICT sector's significant environmental impact. AI technologies, particularly generative models like GPT-4, contribute substantially to global greenhouse gas emissions and water consumption.
The talk assesses how existing and proposed regulations, including EU environmental laws and the GDPR, can be adapted to prioritize sustainability. It advocates for a comprehensive approach to sustainable AI regulation, beyond mere transparency mechanisms for disclosing AI systems' environmental footprint, as proposed in the EU AI Act. The regulatory toolkit must include co-regulation, sustainability-by-design principles, data usage restrictions, and consumption limits, potentially integrating AI into the EU Emissions Trading Scheme. This multidimensional strategy offers a blueprint that can be adapted to other high-emission technologies and infrastructures, such as block chain, the meta-verse, or data centers. Arguably, it is crucial for tackling the twin key transformations of our society: digitization and climate change mitigation.

Institutions

• UHH

Institutions

• UHH

Prof. Dr. Mathias Risse, John F. Kennedy School of Government, Harvard University, Cambridge, MA, USA

Institutions

• UHH

 Prof. Dr. Andra Siibak, University of Tartu, Tartu, Estland

Present day children’s futures are decided by algorithms predicting their probability of success at school, their suitability for a job position, their likely recidivism or mental health problems. Advances in predictive analytics, artificial intelligence (AI) systems, behavioral-, and biometrics technologies, have started to be aggressively used for monitoring, aggregating, and analyzing children’s data. Such dataveillance happening both in homes, schools, and peer networks has a profound impact not only on children’s preferences, social relations, life chances, rights and privacy but also the "future of human agency - and ultimately, of society and culture" (Mascheroni & Siibak 2021: 169).

Building upon the findings of my different empirical case studies, I will showcase how the popular digital parenting practices and the growing datafication happening in the education sector, could create not only hypothetical data scares but also lead to real data scars in the lives of the young.

Institutions

• UHH

Vincent C. Müller is AvH Professor for Philosophy and Ethics of AI and Director of the Centre for Philosophy and AI Research (PAIR) at FAU Erlangen-Nuremberg

It is now frequently observed that there is no proper scope and no proper method in the discipline of AI-ethics. This has become an issue in the development towards maturity of the discipline, e.g. canonical problems, positions, arguments … secure steps forward. We propose a minimal, yet universal view of the field (again Müller 2020). Given this proposal, we will know the scope and the method, and we can appreciate the wide set of contributions.

Institutions

• UHH

Institutions

• UHH

Institutions

• UHH

Institutions

• UHH

Institutions

• UHH

Prof. Dr. Aimee van Wynsberghe, Rheinische Friedrich-Wilhelms-Universität Bonn, D

Institutions

• UHH

• UHH

Institutions

• UHH

Institutions

• UHH

Explore the transformative potential of the Population Dynamics Foundation Model (PDFM), a cutting-edge AI model designed to capture complex, multidimensional interactions among human behaviors, environmental factors, and local contexts. This workshop provides an in-depth introduction to PDFM Embeddings and their applications in geospatial analysis, public health, and socioeconomic modeling. 

Participants will gain hands-on experience with PDFM Embeddings to perform advanced geospatial predictions and analyses while ensuring privacy through the use of aggregated data. Key components of the workshop include: 

• Introduction to PDFM Embeddings: Delve into the model architecture of PDFM and discover how aggregated data (such as search trends, busyness levels, and weather conditions) generates location-specific embeddings.
• Data Preparation: Learn to integrate ground truth data, including health statistics and socioeconomic indicators, with PDFM Embeddings at the postal code or county level.
• Hands-On Exercises: Engage with interactive Colab notebooks to explore real-world applications, such as predicting housing prices using Zillow data and nighttime light predictions with Google Earth Engine data.
• Visualization and Interpretation: Analyze and visualize geospatial predictions and PDFM features in 3D, enhancing your ability to interpret complex datasets. 

By the end of this workshop, participants will have a strong foundation in utilizing PDFM Embeddings to address real-world geospatial challenges. 

Earth system science increasingly relies on machine learning to analyze complex, multivariate, and spatiotemporal data. However, the validity of these models critically depends on the assumption that training and deployment data share similar statistical properties – a condition often violated in real-world environmental applications. This presentation addresses the risks associated with non-stationary training data distributions, arising from climate change, evolving land use, or sensor shifts over time. We show how such distribution shifts can lead to degraded model performance, biased predictions, and misleading scientific conclusions. Through different examples, we illustrate the mechanisms and consequences of non-stationarity. We then discuss methodological solutions, including domain adaptation, continual learning, and uncertainty quantification techniques, that help mitigate these effects and improve model robustness. By combining insights from machine learning and earth system science, this talk aims to foster awareness of distributional risks and promote the development of adaptive, interpretable, and trustworthy models for understanding and predicting Earth’s dynamic systems.

Institutions

• AI for Good

Gastvortrag von Lin Jia, Senior Data Scientists bei Booking.com. Sie wird an unserem Seminar über kausales maschinelles Lernen am 8. Juli 2024 teilnehmen und über vergangene und aktuelle Projekte zu kausaler Inferenz und kausalem maschinellem Lernen bei Booking.com referieren.

Booking.com ist eine der weltweit führenden digitalen Reise-Plattformen und verfügt über ein starkes Team von Datenwissenschaftlern, die Experten sind in der Anwendung und Entwicklung von Methoden für kausale Analysen und maschinelles Lernen in der Industrie.

Über die Referentin: Lin Jia ist eine leitende Datenwissenschaftlerin bei Booking.com. Sie ist spezialisiert auf die Verwendung von kausalen Beobachtungsansätzen zur Bewertung der Auswirkungen von Produktänderungen und leitet die Initiative zur Durchführung robuster und transparenter Kausalanalysen bei Booking.com. Sie wird ihre Erfahrungen aus verschiedenen Projekten zur kausalen Inferenz bei Booking.com teilen.

Der Vortrag ist offen für alle interessierten Forscher:innen und Studierenden, die etwas über die Aktivitäten der Industrie in der kausalen Datenwissenschaft erfahren möchten.

Institutions

• Fakultät für Betriebswirtschaft, Lehrstuhl für Statistik

Gerhard Wellein: Application Knowledge Required: Performance Modeling for Fund and Profit & Axel Klawonn: What can machine learning be used for in domain decomposition methods?

Gerhard Wellein is a Professor for High Performance Computing at the Department for Computer Science of the Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU) and holds a PhD in theoretical physics from the University of Bayreuth. He is a member of the board of directors of the German NHR-Alliance which coordinates the national HPC Tier-2 infrastructures at German universities. As a member of the scientific steering committees of the Leibniz Supercomputing Centre (LRZ) and the Gauss-Centre for Supercomputing (GCS) he is organizing and surveying the compute time application process for national HPC resources. Gerhard Wellein has more than twenty years of experience in teaching HPC techniques to students and scientists from computational science and engineering, is an external trainer in the Partnership for Advanced Computing in Europe (PRACE) and received the “2011 Informatics Europe Curriculum Best Practices Award” (together with Jan Treibig and Georg Hager) for outstanding teaching contributions. His research interests focus on performance modelling and performance engineering, architecture-specific code optimization, novel parallelization approaches and hardware-efficient building blocks for sparse linear algebra and stencil solvers.

Prof. Dr. Axel Klawonn heads the research group on numerical mathematics and scientific computing at the Universität zu Köln. The group works on the development of efficient numerical methods for the simulation of problems from computational science and engineering. This comprises the development of efficient algorithms, their theoretical analysis, and the implementation on large parallel computers with up to several hundreds of thousands of cores. A special focus in the applications is currently on problems from biomechanics/medicine, structural mechanics, and material science. The research is in the field of numerical methods for partial differential equations and high performance parallel scientific computing, including machine learning.

A multitude of ML tasks in particle physics, from unfolding detector effects to refining simulation and extrapolating background estimations, require mapping one arbitrary distribution to another. Several indirect methods have been developed to achieve this, such as classifier-based reweighting on a distribution level, or conditional generative models. However, training an ML model to perform a direct, deterministic mapping has long been a challenging prospect.

In this talk, I introduce the concept of Schrödinger Bridges, ML architecture closely related to Diffusion Models, which enables direct mapping of arbitrary distribution to arbitrary distribution. I demonstrate two implementation approaches with differing upsides and present state-of-the-art results applying Schrödinger Bridges to unfolding and refinement tasks.

Institutions

• UHH, Institut für Experimentalphysik