Scientific Machine Learning (SciML) is a rapidly emerging field in which conventional computational modeling is combined with machine learning techniques. By integrating domain knowledge and machine learning models, SciML has the potential to boost the performance of computational models and to reduce their computational cost. In this symposium, speakers will talk about real-world applications of SciML (and related methods) in diverse fields, ranging from financial risk modeling to weather forecasting and from computational chemistry to power grid control.
Symposium on the Applications of Scientific Machine Learning
In this symposium, speakers will talk about real-world applications of SciML (and related methods) in diverse fields, ranging from financial risk modeling to weather forecasting and from computational chemistry to power grid control.
- 09:00-09:30 Walk-in
- 09:30-09:40 Welcome & Introduction
- 09:40-10:15 Gabrio Rizzuti (Shearwater GeoServices): Unlocking uncertainty quantification for seismic imaging with machine learning
- 10:15-10:50 Jan Viebahn (TenneT): Potential and challenges of AI-powered decision support for power network operations
- 10:50-11:20 Coffee break
- 11:20-12:55 Judith Dijk (TNO): Trustworthy AI for autonomous systems and decision support
- 11:55-12:30 Jonathan Nuttall (Deltares): SciML and Engineering Practice - A question of Acceptance
- 12:30-13:30 Lunch
- 13:30-14:05 Maurice Schmeits (KNMI): Using machine learning to improve weather forecasts: from nowcasting to sub-seasonal forecasting
- 14:05-14:40 Rianne van den Berg (Microsoft Research): AI4Science at Microsoft Research
- 14:40-15:10 Coffee break
- 15:10-15:45 Drona Kandhai (ING and UvA): The potential and challenges ahead for machine learning algorithms in financial risk modeling
- 15:45-16:30 Discussion session
- 16:30-18:00 Drinks
Participation is free of charge but registration is mandatory.
Speakers and abstracts
Abstract: The ability to quantify the uncertainties in seismic imaging is key to many geotechnical applications, such as offshore wind farms or reservoir monitoring for carbon capture and storage (CCS). Alas, large-scale seismic imaging is a challenging inverse problem that requires extreme computing
resources, especially for 3D work, and uncertainty quantification has been decisively out of reach until recently. We discuss promising solutions offered by the recent advances in variational inference and machine learning.
Gabrio Rizzuti is a senior researcher at Shearwater GeoServices, with extensive academic and industry experience in solving inverse problems.
Abstract: Power networks, responsible for transporting electricity across large geographical regions, are complex infrastructures on which modern life critically depends. Correspondingly, power networks represent a backbone of the energy transition. At the same time, increasing renewable energy integration constitutes a historical challenge for system operators when optimizing electricity transportation while avoiding blackouts. More precisely, power network control represents a real-world decision problem that is characterized by large action spaces, sequentiality (including different time horizons), uncertainty, and multiple objectives.
To cope with these challenges, TenneT has launched the Control Room of the Future program to support development of tools when no commercial alternatives are available. In this talk we focus on the potential and challenges of Artificial Intelligence (AI) applied to power system operations, in particular for the development of decision support tools.
Jan Viebahn is a data scientist at TenneT with a focus on developing AI-based decision support tools.
Abstract: AI in general and machine learning are disruptive technologies that promise enormous economic potential in terms of improving the effectiveness and efficiency of products and services for several applications. This presentation focuses on the challenges for applying AI in two types of AI-enabled systems: autonomous systems and decision support systems, e.g. for deploying these systems trustworthy including keeping meaningful human control and preserving privacy.
Examples are shown for an autonomous robot that can be deployed in search and rescue applications, and for decision support systems that provides diabetes-2 advices to both health care professionals and patients.
Judith Dijk is a senior research scientist at TNO, specialised in extracting information from camera images. She is the program manager of TNO’s Applied AI research program.
Abstract: The process of technology acceptance in engineering practice is a nuanced journey, where engineers grapple with both skepticism and wild optimism, all while considering the risk of failure. They assess usability, gauge perceived usefulness, and respond to external pressures, often swayed by both cautious hesitation and enthusiastic anticipation. In this intricate landscape, building trust amidst these diverse attitudes becomes central to fostering genuine technology acceptance and seamless integration within the engineering domain. Here we will explore ways that we can build this trust.
Jonathan Nuttall is a senior researcher at Deltares, where he leads the Data Science pillar.
Abstract: Machine learning has changed weather forecasting research profoundly in the last years. After a short introduction I'll discuss three different examples: (1) nowcasting of precipitation using generative adversarial networks (GANs), (2) post-processing of wind speed forecasts using
convolutional neural networks (CNNs) and (3) sub-seasonal temperature forecasts using a shallow neural network. I'll conclude with an outlook.
Maurice Schmeits is a senior researcher at KNMI and the coordinator of the data science cluster within KNMI's weather and climate modeling department.
Abstract: In July 2022 Microsoft announced a new global team in Microsoft Research that will focus on AI for science, with locations in the UK, China, Germany, the US and the Netherlands. In this talk I will first briefly discuss some of the research areas that we are currently exploring in AI4Science, covering topics such as drug discovery, material generation and neural PDE solvers. Then I will dive a little deeper into recent work that was done at AI4Science, which focuses on the use of score-based generative modeling for coarse-graining (CG) molecular dynamics simulations.
Rianne van den Berg is a Principal Researcher at Microsoft Research Amsterdam, where she works on the intersection of deep learning and computational chemistry and physics for molecular simulation.
Abstract: In this talk I will first give a brief introduction to the field of financial risk modeling in the context of Financial Derivatives. Some progress related to the application of ML algorithms in combination with stochastic modeling will be discussed. The presentation will end with a brief discussion of challenges ahead.
Drona Kandhai is the head of Quantitative Analytics at ING, and professor in Computational Finance at the University of Amsterdam. He is the Scientific Director of the AI4FinTech initiative of the UvA.