A Calculus for Stochastic Interventions:Causal Effect Identification and Surrogate Experiments

Correa, Juan D.; Bareinboim, Elias

doi:10.1609/aaai.v34i06.6567

Cited by 29 publications

(36 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…A more general framework, named soft interventions, assumes that parent-child dependencies are ''modified'' but yet preserved after intervention. In this setting, Correa and Bareinboim (2020) introduce a set of rules (named r-calculus) for the identifiability of causal effects arising from soft interventions. They then show how these rules can be applied to identify the causal effect of an interventions from a combination of observational and interventional data, the latter arising from exogenous perturbations of selected variables in the system.…”

Section: Discussionmentioning

confidence: 99%

Structural learning and estimation of joint causal effects among network-dependent variables

Castelletti

Mascaro

2021

Stat Methods Appl

View full text Add to dashboard Cite

Bayesian networks in the form of Directed Acyclic Graphs (DAGs) represent an effective tool for modeling and inferring dependence relations among variables, a process known as structural learning. In addition, when equipped with the notion of intervention, a causal DAG model can be adopted to quantify the causal effect on a response due to a hypothetical intervention on some variable. Observational data cannot distinguish between DAGs encoding the same set of conditional independencies (Markov equivalent DAGs), which however can be different from a causal perspective. In addition, because causal effects depend on the underlying network structure, uncertainty around the DAG generating model crucially affects the causal estimation results. We propose a Bayesian methodology which combines structural learning of Gaussian DAG models and inference of causal effects as arising from simultaneous interventions on any given set of variables in the system. Our approach fully accounts for the uncertainty around both the network structure and causal relationships through a joint posterior distribution over DAGs, DAG parameters and then causal effects.

show abstract

Section: Discussionmentioning

confidence: 99%

Structural learning and estimation of joint causal effects among network-dependent variables

Castelletti

Mascaro

2021

Stat Methods Appl

View full text Add to dashboard Cite

show abstract

“…Further work includes drawing connections with other research programs, such as questions related to identification causal structure [19,20,22] or extensions of do-calculus [7]. As argued in Sect.…”

Section: Discussionmentioning

confidence: 99%

“…Causal graphical models move the focus from joint probability distributions to functional dependencies thanks to the Structural Causal Model (SCM) framework. Several extensions to this do-calculus have been proposed recently [25,14,23,7]. Pearl's seminal paper supposes a Directed Acyclic Graph (DAG) structure.…”

Section: Introductionmentioning

confidence: 99%

Causal Inference Theory with Information Dependency Models

Heymann,

de Lara,

Chancelier

2021

Preprint

View full text Add to dashboard Cite

Inferring the potential consequences of an unobserved event is a fundamental scientific question. To this end, Pearl's celebrated do-calculus provides a set of inference rules to derive an interventional probability from an observational one. In this framework, the primitive causal relations are encoded as functional dependencies in a Structural Causal Model (SCM), which are generally mapped into a Directed Acyclic Graph (DAG) in the absence of cycles. In this paper, by contrast, we capture causality without reference to graphs or functional dependencies, but with information fields and Witsenhausen's intrinsic model. The three rules of do-calculus reduce to a unique sufficient condition for conditional independence, the topological separation, which presents interesting theoretical and practical advantages over the d-separation. With this unique rule, we can deal with systems that cannot be represented with DAGs, for instance systems with cycles and/or 'spurious' edges. We treat an example that cannot be handled -to the extent of our knowledge -with the tools of the current literature. We also explain why, in the presence of cycles, the theory of causal inference might require different tools, depending on whether the random variables are discrete or continuous.

show abstract

“…In light of the causal graph, we can define valid interventions (Pearl, 2009;Schölkopf et al, 2021). Here we consider both hard and soft interventions (Eberhardt & Scheines, 2007;Correa & Bareinboim, 2020) on the perturbation variable E. Each kind of intervention on E would give us a distribution, which could be either natural or adversarial, over the observed variables. Without loss of generality, we assume the model without any intervention represents a natural perturbation process.…”

Section: A Causal View On Adversarial Data Generationmentioning

confidence: 99%

CausalAdv: Adversarial Robustness through the Lens of Causality

Zhang¹,

Gong²,

Liu³

et al. 2021

Preprint

View full text Add to dashboard Cite

The adversarial vulnerability of deep neural networks has attracted significant attention in machine learning. From a causal viewpoint, adversarial attacks can be considered as a specific type of distribution change on natural data. As causal reasoning has an instinct for modeling distribution change, we propose to incorporate causality into mitigating adversarial vulnerability. However, causal formulations of the intuition of adversarial attack and the development of robust DNNs are still lacking in the literature. To bridge this gap, we construct a causal graph to model the generation process of adversarial examples and define the adversarial distribution to formalize the intuition of adversarial attacks. From a causal perspective, we find that the label is spuriously correlated with the style (content-independent) information when an instance is given. The spurious correlation implies that the adversarial distribution is constructed via making the statistical conditional association between style information and labels drastically different from that in natural distribution. Thus, DNNs that fit the spurious correlation are vulnerable to the adversarial distribution. Inspired by the observation, we propose the adversarial distribution alignment method to eliminate the difference between the natural distribution and the adversarial distribution. Extensive experiments demonstrate the efficacy of the proposed method. Our method can be seen as the first attempt to leverage causality for mitigating adversarial vulnerability.

show abstract

A Calculus for Stochastic Interventions:Causal Effect Identification and Surrogate Experiments

Cited by 29 publications

References 14 publications

Structural learning and estimation of joint causal effects among network-dependent variables

Structural learning and estimation of joint causal effects among network-dependent variables

Causal Inference Theory with Information Dependency Models

CausalAdv: Adversarial Robustness through the Lens of Causality

Contact Info

Product

Resources

About