Past Meetings

TimeProgram
14:30-15:15Saber Salehkaleybar (Leiden University) - Causal Inference in Linear Systems: From Acyclic Models to Feedback Dynamics.

Linear systems are prevalent across science and engineering, and often provide interpretable first-order models of more complex phenomena. In this talk, I will present recent identifiability results for causal inference in linear systems, moving from acyclic structural causal models (SCMs) to cyclic ones and continuous-time stochastic dynamics. I will first discuss how non-Gaussianity and higher-order moments enable causal effect identification in acyclic linear SCMs with latent confounding, including proxy, instrumental-variable, and heterogeneous-environment settings. I will then turn to feedback systems. Cyclic linear SCMs provide one way to model feedback, and I will discuss distributional equivalence in cyclic linear non-Gaussian models. Finally, I will consider linear stochastic differential equations as another framework for modeling feedback through continuous-time interactions. When only the stationary distribution is observed, exact causal-effect strengths are generally not identifiable without additional assumptions. I will instead focus on edge-sign identifiability, which asks when the sign of a direct causal effect is uniquely determined by the stationary covariance matrix. I will present criteria for determining whether an edge sign is identifiable, non-identifiable, or partially identifiable.
15:15-15:30Break
15:30-16:15Junhyung Park (ETH Zürich) - Causal Spaces: A Measure-Theoretic Axiomatisation of Causality.

Mathematical theories are founded upon simple axioms that isolate the essential structure of an object or a concept. Abstract algebra starts with the group axioms, linear algebra the vector space axioms, and topology the axioms of open sets. Most relevantly, Kolmogorov's axioms of probabilities based on measure theory have established probability theory as a branch of pure mathematics in the last century. The late 20th century saw the introduction of two hugely influential mathematical frameworks of causality, namely, the structural causal models of Judea Pearl and the potential outcomes of Donald Rubin. Their value, especially in causal inference from data, have been proved beyond doubt. However, these, and other frameworks of causality, were not designed as minimal axiomatic foundations, in the spirit of the above mathematical theories. To fill this gap, we introduce causal spaces, with two minimal axioms capturing the essence of interventions, namely, (i) doing nothing changes nothing (trivial intervention), and (ii) when X is given value x, X has value x (interventional determinism). We show that the prior formalisms can be recovered as special cases of causal spaces, and that a rich and fruitful mathematical theory, with clear causal semantics, can arise from these minimal axioms.
16:30-17:30Drinks at Grand Café Living (in the same building)
TimeProgram
14:30-14:45Opening
14:45-15:30Eni Musta (University of Amsterdam) - Can we detect treatment effect waning from time-to-event data? (abstract)
15:30-15:45Break
15:45-16:30Hidde Fokkema (University of Amsterdam) Sample-efficient Learning of Concepts with Theoretical Guarantees: from Data to Concepts without Interventions. (abstract)
16:30-17:30Drinks at Polder
TimeProgram
14:30-14:45Tom Claassen (Radboud University Nijmegen) - Sound and complete causal inference with background knowledge in the presence of latent confounders and selection bias. (abstract)
15:30-15:45Onno Zoeter (Booking.com) - When the problem becomes richer than supervised learning. A real-world use of causality in machine learning. (abstract)
16:30-17:30Drinks
TimeProgram
14:15-15:00Rickard Karlsson (TU Delft) - Robust integration of external control data in randomized trials (abstract)
15:15-16:00Sourbh Bhadane (UvA) - Revisiting the Berkeley Admissions data: Statistical Tests for Causal Hypotheses (abstract)
16:00-17:00Drinks
TimeProgram
10:00-11:00Thomas Richardson (University of Washington) - Statistical analysis for the discrete instrumental variable model (abstract)
TimeProgram
14:30-15:30Riccardo Massidda (University of Pisa) - Frameworks for Representing and Learning Abstract Causal Models (abstract and slides)
15:30-16:30Phillip Lippe (UvA) - Causal Representation Learning across Multiple Environments (abstract and slides)
16:30-17:30Drinks
TimeProgram
14:30-15:30Hessel Peters-Sengers (AUMC) - ICU-acquired complications (abstract and slides)
15:30-16:30Richard Post (TU/e) - Flexible Machine Learning Estimation of Conditional Average Treatment Effects (abstract and slides)
16:30-17:30Drinks
TimeProgram
14:30-15:30Mirthe van Diepen (Radboud University) - Detecting New Risk Factors via Causal Discovery in Aortic Surgery (abstract)
15:30-16:30Philip Boeken (UvA) - A Markov property for sample paths of stochastic processes (abstract and slides)
16:30-18:00Drinks
TimeProgram
15:00-16:00Johannes Textor (RU, RUMC) - Are DAGs Really Useful for Causal Reasoning in Complex Systems? (abstract)
16:00-17:00Julia Kowalska (AUMC) - Regression Discontinuity Design from Bayesian perspective: opportunities and challenges (abstract and slides)
17:00-18:30Drinks
TimeProgram
15:00-16:00Nan van Geloven (LUMC) - Prediction under hypothetical interventions: evaluation of counterfactual performance using longitudinal observational data (abstract and slides)
16:00-17:00Sander Beckers (UvA) - Moral responsibility for AI systems (abstract and slides)
17:00-18:30Drinks