Fwd: Van Dantzig seminar (May 19)
4 views
Skip to first unread message
p.a.boeken
Apr 16, 2026, 10:13:41 AM (13 days ago) Apr 16
to Amsterdam Causality Meeting
Dear all,
On behalf of Eni Musta I would like to draw attention to the Van Dantzig seminar on May 19th, where Peter Bühlmann will give a presentation titled "Causality-Inspired Prediction: A Path to more Robust Algorithms”. See the message below.
Best,
Philip
————————————————————
Dear all,
The next van Dantzig seminar will take place on Tuesday, 19 May, at the University of Amsterdam, Science Park 904. The speakers will be Peter Bühlmann (ETH Zurich) and Almut Veraart (Imperial College London). There will be two introductory talks in the morning and two more in-depth talks in the afternoon.
The programme is as follows:
Morning session (lecture room G0.05)
10:00-11.00 Peter Bühlmann, "Domain Generalization and Adaptation in Digital Health (and beyond)"
11:00-11:15 Coffee break
11:15-12:15 Almut Veraart, "An introduction to ambit stochastics"
12:15-14:00 Lunch break
Afternoon session (lecture room A1.16)
14:00-15:00 Peter Bühlmann, "Causality-Inspired Prediction: A Path to more Robust Algorithms"
15:00-15:15 Coffee break
15:15-16:15 Almut Veraart, "Nonparametric estimation of trawl processes: Theory and applications to high-frequency financial data"
16:15- Drinks (Polder)
The abstracts of the talks can be found below. For more information about the Van Dantzig seminar, please visit the website: https://www.vvsor.nl/mathematical-statistics/pages/van-dantzig-seminar/
Everybody is cordially invited to attend. We look forward to meeting you there.
Best wishes,
Hanne and Eni
Abstracts
Domain Generalization and Adaptation in Digital Health (and beyond) - P. Bühlmann
Statistical models and machine learning algorithms are often deployed in populations that differ from those on which they were trained, a challenge that is particularly acute in digital health. We discuss domain generalization and adaptation for a large-scale database from multiple countries with intensive care unit (ICU) data. We introduce Distributionally Robust Invariance Learning as an approach to exploiting stable structure across environments, and conclude with a brief discussion of the potential and limitations of a novel foundation models in this context.
Causality-Inspired Prediction: A Path to more Robust Algorithms - P. Bühlmann
This talk examines in greater detail how ideas from causality can inform the construction of predictive procedures that remain reliable under distributional change. We highlight connections between invariance, distributional robustness, and underlying causal structure, and discuss their implications for transportability across environments. The theme is also closely related to the problem of extrapolation beyond the support of the observed data, which we will briefly touch upon.
An introduction to ambit stochastics - A. Veraart
What do turbulent fluid flows, electricity price spikes, and wind capacity factors across a power grid have in common? All can be modelled within the framework of ambit stochastics, a probabilistic language for spatio-temporal phenomena built on stochastic integrals over moving space-time regions. This lecture provides a self-contained introduction to the theory, covering ambit sets, Lévy bases, ambit fields, and their purely temporal specialisation to Lévy and Brownian semistationary processes. A central theme is that the kernel function acts as a direct handle on the memory and correlation structure of the process, giving the modeller precise and interpretable control. We illustrate this through applications to electricity spot and forward price modelling, and conclude with a forward look at how the framework scales to modern high-dimensional network data.
Nonparametric estimation of trawl processes: Theory and applications to high-frequency financial data - A. Veraart
How much does the past matter, and for how long? For count-valued time series, such as trade counts, bid-ask spreads, order arrivals, this question is both practically important and surprisingly hard to answer without imposing strong parametric assumptions. Trawl processes offer a continuous-time framework in which the memory structure is encoded by a single function, the trawl function, which can be estimated directly and nonparametrically from data. We introduce this class of processes, establish the key identity connecting the trawl function to the autocovariance of the observed series, and develop a complete inferential framework including consistency, central limit theorems across three asymptotic regimes, bias correction, and feasible confidence bands. We show how these tools can be used to diagnose misspecification of standard parametric models, and how the nonparametric estimator translates directly into an optimal forecasting rule. The methodology is illustrated on limit order book data from the NYSE, where it delivers measurable forecasting improvements over exponential and other parametric trawl specifications.
On behalf of Eni Musta I would like to draw attention to the Van Dantzig seminar on May 19th, where Peter Bühlmann will give a presentation titled "Causality-Inspired Prediction: A Path to more Robust Algorithms”. See the message below.
Best,
Philip
————————————————————
Dear all,
The next van Dantzig seminar will take place on Tuesday, 19 May, at the University of Amsterdam, Science Park 904. The speakers will be Peter Bühlmann (ETH Zurich) and Almut Veraart (Imperial College London). There will be two introductory talks in the morning and two more in-depth talks in the afternoon.
The programme is as follows:
Morning session (lecture room G0.05)
10:00-11.00 Peter Bühlmann, "Domain Generalization and Adaptation in Digital Health (and beyond)"
11:00-11:15 Coffee break
11:15-12:15 Almut Veraart, "An introduction to ambit stochastics"
12:15-14:00 Lunch break
Afternoon session (lecture room A1.16)
14:00-15:00 Peter Bühlmann, "Causality-Inspired Prediction: A Path to more Robust Algorithms"
15:00-15:15 Coffee break
15:15-16:15 Almut Veraart, "Nonparametric estimation of trawl processes: Theory and applications to high-frequency financial data"
16:15- Drinks (Polder)
The abstracts of the talks can be found below. For more information about the Van Dantzig seminar, please visit the website: https://www.vvsor.nl/mathematical-statistics/pages/van-dantzig-seminar/
Everybody is cordially invited to attend. We look forward to meeting you there.
Best wishes,
Hanne and Eni
Abstracts
Domain Generalization and Adaptation in Digital Health (and beyond) - P. Bühlmann
Statistical models and machine learning algorithms are often deployed in populations that differ from those on which they were trained, a challenge that is particularly acute in digital health. We discuss domain generalization and adaptation for a large-scale database from multiple countries with intensive care unit (ICU) data. We introduce Distributionally Robust Invariance Learning as an approach to exploiting stable structure across environments, and conclude with a brief discussion of the potential and limitations of a novel foundation models in this context.
Causality-Inspired Prediction: A Path to more Robust Algorithms - P. Bühlmann
This talk examines in greater detail how ideas from causality can inform the construction of predictive procedures that remain reliable under distributional change. We highlight connections between invariance, distributional robustness, and underlying causal structure, and discuss their implications for transportability across environments. The theme is also closely related to the problem of extrapolation beyond the support of the observed data, which we will briefly touch upon.
An introduction to ambit stochastics - A. Veraart
What do turbulent fluid flows, electricity price spikes, and wind capacity factors across a power grid have in common? All can be modelled within the framework of ambit stochastics, a probabilistic language for spatio-temporal phenomena built on stochastic integrals over moving space-time regions. This lecture provides a self-contained introduction to the theory, covering ambit sets, Lévy bases, ambit fields, and their purely temporal specialisation to Lévy and Brownian semistationary processes. A central theme is that the kernel function acts as a direct handle on the memory and correlation structure of the process, giving the modeller precise and interpretable control. We illustrate this through applications to electricity spot and forward price modelling, and conclude with a forward look at how the framework scales to modern high-dimensional network data.
Nonparametric estimation of trawl processes: Theory and applications to high-frequency financial data - A. Veraart
How much does the past matter, and for how long? For count-valued time series, such as trade counts, bid-ask spreads, order arrivals, this question is both practically important and surprisingly hard to answer without imposing strong parametric assumptions. Trawl processes offer a continuous-time framework in which the memory structure is encoded by a single function, the trawl function, which can be estimated directly and nonparametrically from data. We introduce this class of processes, establish the key identity connecting the trawl function to the autocovariance of the observed series, and develop a complete inferential framework including consistency, central limit theorems across three asymptotic regimes, bias correction, and feasible confidence bands. We show how these tools can be used to diagnose misspecification of standard parametric models, and how the nonparametric estimator translates directly into an optimal forecasting rule. The methodology is illustrated on limit order book data from the NYSE, where it delivers measurable forecasting improvements over exponential and other parametric trawl specifications.
Reply all
Reply to author
Forward
0 new messages