Research.

Preprints.

A Topological Sorting Criterion for Random Causal Directed Acyclic Graphs

AG Reisach, A Chambaz, G Blanchard, S Weichwald

Common random DAG generators induce increasing numbers of relatives along the causal order, enabling order recovery and exposing structural biases in synthetic discovery benchmarks.

pdf arXiv

Identifying Causal Effects Using a Single Proxy Variable

S Vollmer, N Pfister^*, S Weichwald^*; ^*Equal contribution

Under suitable completeness conditions, a single proxy suffices to identify causal effects despite unobserved confounding; introduces SPICE and a neural network estimator for effect estimation.

pdf arXiv

Linear-Time Primitives for Algorithm Development in Graphical Causal Inference

M Wienöbst, S Weichwald, L Henckel

CIfly reframes graphical causal inference tasks as reachability queries in purpose-built graphs. The declarative framework enables scalable algorithm design, has a fast Rust core, and is available in Python (pip install ciflypy) and R (install.packages("ciflyr")).

pdf arXiv code documentation

What is causal about causal models and representations?

FH Jørgensen, L Gresele, S Weichwald

Formalises the requirements for interpreting real-world actions as interventions in causal models. A natural interpretation of actions as interventions renders intervention-based falsification circular, and no interpretation can be both non-circular and satisfy a set of natural desiderata.

pdf arXiv

Peer-reviewed.

The Case for Time in Causal DAGs

AG Reisach, A Suárez, S Weichwald, A Chambaz
Philosophy of Science (accepted)

Argues that causal variables require temporal qualification and formalises how this sharpens the interpretation of causal DAGs and the acyclicity assumption.

pdf arXiv

Unfair Utilities and First Steps Towards Improving Them

FH Jørgensen, S Weichwald, J Peters
Journal of Causal Inference (accepted)

Reframes fairness from constraints on policies or predictors to conditions on utilities; utilities that incentivise reconstructing protected attributes can be unfair even when only essential features are used.

pdf arXiv

Embracing Discrete Search: A Reasonable Approach to Causal Structure Learning

M Wienöbst, L Henckel, S Weichwald
ICLR, 2026

Revisits discrete search for causal structure learning and shows that careful order-and-parent search can be practical at scale, with strong empirical performance across benchmarks. Available in Python (pip install flopsearch) and R.

pdf OpenReview arXiv code

Causal Foundations of Collective Agency

FH Jørgensen, S Weichwald, L Hammond
CLeaR, 2026

Formalises when groups of agents can be treated as a single rational entity, using causal games and abstraction to analyse emergent collective agency.

pdf OpenReview arXiv

All or None: Identifiable Linear Properties of Next-token Predictors in Language Modeling

E Marconato, S Lachapelle, S Weichwald^*, L Gresele^*; ^*Equal contribution
AISTATS, 2025

Under suitable conditions, linear properties either hold in all or none distribution-equivalent next-token predictors.

pdf arXiv

Identifying Causal Effects using Instrumental Time Series: Nuisance IV and Correcting for the Past

N Thams, R Søndergaard, S Weichwald, J Peters
Journal of Machine Learning Research, 2024

Develops graphical identification theory for instrumental-variable methods in time series. Shows why adjusting for past states is needed for valid instruments and introduces Nuisance IV, a modified IV estimator.

pdf JMLR arXiv code

Adjustment Identification Distance: A gadjid for Causal Structure Learning

L Henckel, T Würtzen, S Weichwald
UAI, 2024

Introduces task-oriented causal graph distances based on downstream adjustment-identification behaviour, together with polynomial-time algorithms, including the first such distance for CPDAGs. Efficient implementations are available for Python (pip install gadjid) and R (install.packages("gadjid")).

pdf arXiv code

spillR: Spillover Compensation in Mass Cytometry Data

M Guazzini, AG Reisach, S Weichwald, C Seiler
Bioinformatics, 2024

Introduces a nonparametric finite-mixture model for spillover compensation in mass cytometry, transferring spillover distributions estimated from beads to real data.

pdf doi bioRxiv R package article .Rmd

A Scale-Invariant Sorting Criterion to Find a Causal Order in Additive Noise Models

AG Reisach, M Tami, C Seiler, A Chambaz, S Weichwald
NeurIPS, 2023

Introduces R²-sortability, a scale-invariant signal for causal ordering in additive noise models. Unlike var-sortability, it is preserved under standardisation and rescaling, and can still be exploited to learn causal structure.

pdf OpenReview arXiv CausalDisco

Learning by Doing: Controlling a Dynamical System using Causality, Control, and Reinforcement Learning

S Weichwald, SW Mogensen, TE Lee, D Baumann, O Kroemer, I Guyon, S Trimpe, J Peters, N Pfister
NeurIPS 2021 Competitions and Demonstrations Track, 2022

Describes a NeurIPS competition connecting causality, control, and reinforcement learning around the question of how systems change when we interact with them.

pdf arXiv code Track CHEM Track ROBO LBD competition

Beware of the Simulated DAG! Causal Discovery Benchmarks May Be Easy To Game

AG Reisach, C Seiler, S Weichwald
NeurIPS, 2021

In additive noise models (ANMs), marginal variances often reveal causal order, making simulated benchmarks easy to game. We introduce var-sortability as a diagnostic for this pattern and argue that causal-discovery benchmarks should report it.

pdf OpenReview arXiv code explainer

Compositional Abstraction Error and a Category of Causal Models

EF Rischel, S Weichwald
UAI, 2021

Develops a categorical framework for finite interventional causal models and their abstraction errors, showing that compositionality can be achieved for causal model transformations and their associated errors.

pdf arXiv

Improving 1-year mortality prediction in ACS patients using machine learning

S Weichwald, A Candreva^*, R Burkholz^*, R Klingenberg, L Räber, D Heg, R Manka, B Gencer, F Mach, D Nanchen, N Rodondi, S Windecker, R Laaksonen, SL Hazen, A Eckardstein, F Ruschitzka, TF Lüscher, JM Buhmann, CM Matter; ^*Equal contribution
European Heart Journal: Acute Cardiovascular Care, 2021

Studies machine-learning model development for small, imbalanced clinical data in one-year mortality prediction after acute coronary syndrome.

pdf doi free access

Causality in Cognitive Neuroscience: Concepts, Challenges, and Distributional Robustness

S Weichwald, J Peters
Journal of Cognitive Neuroscience, 2021

Explains how distributional robustness and generalisability can guide causality research in cognitive neuroscience, especially when interventional data are scarce and the right variables are difficult to define. Also provides an accessible introduction to selected causal discovery ideas, assumptions, and limitations.

pdf doi arXiv

Causal structure learning from time series: Large regression coefficients may predict causal links better in practice than small p-values

S Weichwald, ME Jakobsen, PB Mogensen, L Petersen, N Thams, G Varando
NeurIPS 2019 Competition and Demonstration Track, 2020

Describes the competition-winning algorithms for causal structure learning from time series in the NeurIPS Causality4Climate challenge. Examines why large regression coefficients may outperform small p-values as causal-link signals and why normalisation can hinder discovery.

pdf arXiv tidybench

Robustifying Independent Component Analysis by Adjusting for Group-Wise Stationary Noise

N Pfister^*, S Weichwald^*, P Bühlmann, B Schölkopf; ^*Equal contribution
Journal of Machine Learning Research, 2019

Develops coroICA, a confounding-robust ICA method that accounts for group-wise stationary noise and provides identifiable source recovery under heterogeneous confounding.

pdf JMLR arXiv Python/R/matlab audible example EEG example video

Personalized brain-computer interface models for motor rehabilitation

AA Mastakouri, S Weichwald, O Özdenizci, T Meyer, B Schölkopf, M Grosse-Wentrup
IEEE International Conference on Systems, Man, and Cybernetics, 2017

Proposes combining brain-computer interface training and transcranial electrical stimulation as complementary routes towards personalised models for motor rehabilitation.

pdf doi arXiv

Causal Consistency of Structural Equation Models

PK Rubenstein^*, S Weichwald^*, S Bongers, JM Mooij, D Janzing, M Grosse-Wentrup, B Schölkopf; ^*Equal contribution
UAI, 2017

Formalises when structural causal models at different levels of description are consistent, in the sense that transformations between models preserve interventional conclusions.

pdf arXiv

Absence of EEG correlates of self-referential processing depth in ALS

T Fomina, S Weichwald, M Synofzik, J Just, L Schöls, B Schölkopf, M Grosse-Wentrup
PLoS ONE, 2017

Finds EEG correlates (bandpower modulation in medial prefrontal cortex) of self-referential processing in healthy individuals but not in people with ALS.

pdf doi

MERLiN: Mixture Effect Recovery in Linear Networks

S Weichwald, M Grosse-Wentrup, A Gretton
IEEE Journal of Selected Topics in Signal Processing, 2016

Introduces MERLiN, a causal inference algorithm for recovering a causal effect variable from an observed linear mixture, robust to hidden confounding.

pdf doi arXiv code

Pymanopt: A Python Toolbox for Optimization on Manifolds using Automatic Differentiation

J Townsend, N Koep, S Weichwald
Journal of Machine Learning Research, 2016

Introduces Pymanopt, a Python toolbox for optimisation on manifolds using automatic differentiation, lowering the implementation barrier for manifold optimisation methods. Example: manifold optimisation for inferring parameters of a MoG model.

pdf JMLR arXiv code documentation

Recovery of non-linear cause-effect relationships from linearly mixed neuroimaging data

S Weichwald, A Gretton, B Schölkopf, M Grosse-Wentrup
International Workshop on Pattern Recognition in Neuroimaging, 2016

Extends MERLiN to recover non-linear cause-effect relationships from linearly mixed neuroimaging data.

pdf doi arXiv code slides

Causal interpretation rules for encoding and decoding models in neuroimaging

S Weichwald, T Meyer, O Özdenizci, B Schölkopf, T Ball, M Grosse-Wentrup
NeuroImage, 2015

Clarifies which causal interpretations are warranted for encoding and decoding models in neuroimaging, and how combining the two can support additional causal insights.

pdf doi arXiv explainer slides

Causal and anti-causal learning in pattern recognition for neuroimaging

S Weichwald, B Schölkopf, T Ball, M Grosse-Wentrup
International Workshop on Pattern Recognition in Neuroimaging, 2014

Argues that it is not sufficient to distinguish between encoding and decoding models in neuroimaging; the interpretation of such models also depends on whether they are employed in a stimulus- or response-based setting.

pdf doi arXiv

Decoding index finger position from EEG using random forests

S Weichwald, T Meyer, B Schölkopf, T Ball, M Grosse-Wentrup
International Workshop on Cognitive Information Processing, 2014

Shows that index finger positions can be decoded from non-invasive EEG recordings, with high β-power over contralateral sensorimotor cortex carrying most information. Best student paper award.

pdf doi arXiv

Thesis.

Pragmatism and Variable Transformations in Causal Modelling

S Weichwald
ETH Zurich, 2019

The statistical treatment of causal modelling lays out methodology that, under well-specified assumptions, enables us to infer cause-effect relationships from observational data. The adoption and fruitful utilisation of such methods remains limited, however, despite the statistical foundations and numerous theoretical advances. In this thesis, we present contributions towards closing the gap between statistical causal modelling and its successful application.

pdf doi

Other.

What is Cantor’s continuum problem?

S Weichwald
2013

This seminar paper reviews Kurt Gödel’s article »What is Cantor’s continuum problem?« with short recaps, rough explanations, and selective examples to make the argument largely self-contained.

pdf

Software.

CausalDisco: Python toolbox for SortnRegress baselines and R²-/Var-sortability diagnostics for causal discovery.
code documentation
CIfly: Declarative framework for designing efficient graphical causal inference algorithms.
code documentation
coroICA: Confounding-robust independent component analysis.
Python R matlab documentation
flopsearch: Causal discovery algorithm embracing discrete search.
code
gadjid: Adjustment identification distances for comparing DAGs and CPDAGs.
code
MERLiN: Mixture effect recovery for reconstructing causal variables from linear mixtures.
code
Pymanopt: Python toolbox for optimisation on manifolds with automatic differentiation.
Part of the manopt family.
code documentation
tidybench: Implementation of competition-winning causal structure learning algorithms for time series.
code