• Corpus ID: 245117934

Learning soft interventions in complex equilibrium systems

  title={Learning soft interventions in complex equilibrium systems},
  author={Michel Besserve and Bernhard Sch{\"o}lkopf},
  booktitle={Conference on Uncertainty in Artificial Intelligence},
Complex systems often contain feedback loops that can be described as cyclic causal models. Intervening in such systems may lead to counterintuitive effects, which cannot be inferred directly from the graph structure. After establishing a framework for differentiable soft interventions based on Lie groups, we take advantage of modern automatic differentiation techniques and their application to implicit functions in order to optimize interventions in cyclic causal models. We illustrate the use… 

Figures from this paper



BACKSHIFT: Learning causal cyclic graphs from unknown shift interventions

A simple method to learn linear causal cyclic models in the presence of latent variables that only uses second moments of the data and performs simple joint matrix diagonalization, applied to differences between covariance matrices.

Causal Models for Dynamical Systems

This chapter introduces two types of causal kinetic models that differ in how the randomness enters into the model: it may either be considered as observational noise or as systematic driving noise, and provides a possible starting point for causal inference.

From Ordinary Differential Equations to Structural Causal Models: the deterministic case

We show how, and under which conditions, the equilibrium states of a first-order Ordinary Differential Equation (ODE) system can be described with a deterministic Structural Causal Model (SCM). Our

Conditional Independences and Causal Relations implied by Sets of Equations

It is proved that, under a solvability assumption, a Markov ordering graph can be constructed that implies conditional independences and a causal ordering graph that encodes the effects of soft and perfect interventions by making use of Simon's causal ordering algorithm.

Causality and independence in perfectly adapted dynamical systems

It is demonstrated that perfect adaptation can lead to misleading orientation of edges in the output of causal discovery algorithms and is used to provide a sufficient graphical condition for a dynamical system to achieve perfect adaptation, which does not require simulations or cumbersome calculations.

Interventions and Causal Inference

The literature on causal discovery has focused on interventions that involve randomly assigning values to a single variable. But such a randomized intervention is not the only possibility, nor is it

Causal Discovery from Soft Interventions with Unknown Targets: Characterization and Learning

This paper investigates the task of structural learning in non-Markovian systems from a combination of observational and soft experimental data when the interventional targets are unknown and develops an algorithm capable of harnessing the collection of data to learn the corresponding equivalence class.

Characterization and Learning of Causal Graphs with Latent Variables from Soft Interventions

The FCI algorithm is extended to combine observational and interventional datasets, including new orientation rules particular to this setting, and a novel notion of interventional equivalence class of causal graphs with latent variables based on these invariances is introduced.

Foundations of structural causal models with cycles and latent variables

The foundations for a general theory of statistical causal modeling with SCMs are provided, allowing for the presence of both latent confounders and cycles, and a class of simple SCMs is introduced that extends the class of acyclic SCMs to the cyclic setting, while preserving many of the convenient properties.

General Transportability of Soft Interventions: Completeness Results

A graphical condition is developed that is both necessary and sufficient for deciding soft-transportability and an algorithm is developed to determine whether a non-atomic intervention is computable from a combination of the distributions available across domains.