Relaxing Faithfulness with Intervention-Only Causal Discovery

Noosaga Placements

• Machine Learningsubfield90%
  The paper focuses on causal discovery algorithms, a core topic in machine learning. It introduces new assumptions and algorithms for learning causal structures from data, which is a fundamental machine learning problem. The paper is categorized under cs.LG and stat.ML.
  Causal discovery algorithms learn a network that describes the causal dependencies among random variables.In this paper, we argue that hard interventions contain information about the presence/absence of causal linkage that is overlooked in the first stage of structure discovery.We show that a mild assumption — called intervention-immediacy faithfulness — that allows cancellations, is sufficient to nonparametrically identify causal structures with hard interventions.
• Bayesian Networksframework90%
  The paper is situated within the framework of Bayesian Networks, as it deals with causal discovery of DAGs, which are a type of Bayesian network. The paper discusses the causal Markov condition, d-separation, and faithfulness, all of which are key concepts in Bayesian networks.
  Structural Causal Models (SCMs), popularized by the works of Pearl [1998, 2009], graphically describe networks of causal dependencies.Constraintbased approaches use the observation that variables without a direct causal link can be made independent by conditioning on variables on intermediary causal paths, known as the causal Markov condition [Pearl, 2009].D-separation provides a graphical criterion that constitutes a necessary but not sufficient condition for conditional independence under the causal Markov condition.
• Statistical Learningframework80%
  The paper is situated within the Statistical Learning framework, as it deals with learning causal structures from data using statistical tests and assumptions. It provides a parametric analysis of faithfulness assumptions and uses statistical tests for change detection.
  A critical assumption for the first step is faithfulness: a requirement that causally linked variables exhibit statistical dependence.To provide intuition about path cancellation, we give an example with linear Gaussian structural equation models (SEMs).We instantiate intervention-based causal discovery with a two-sample Student-t test (scipy.stats, significance level 0.05, Bonferroni-corrected for the number of tests) to detect a change in mean
• Statistical Learningsubfield70%
  The paper deals with statistical assumptions (faithfulness) and uses statistical tests (two-sample t-test) for change detection. It also provides a parametric analysis in the Gaussian setting, comparing the volume of violations of different faithfulness assumptions. This connects to statistical learning theory.
  A critical assumption for the first step is faithfulness: a requirement that causally linked variables exhibit statistical dependence.To provide intuition about path cancellation, we give an example with linear Gaussian structural equation models (SEMs).We instantiate intervention-based causal discovery with a two-sample Student-t test (scipy.stats, significance level 0.05, Bonferroni-corrected for the number of tests) to detect a change in mean
• Probabilistic Graphical Modelsframework70%
  The paper is situated within the Probabilistic Graphical Models framework, as it deals with causal discovery, which is a key application of probabilistic graphical models. The paper discusses structural causal models, which are a type of probabilistic graphical model.
  Structural Causal Models (SCMs), popularized by the works of Pearl [1998, 2009], graphically describe networks of causal dependencies.Causal discovery is the task of recovering the underlying causal structure from data, typically in the form of a directed acyclic graph (DAG) or a representation of an equivalence class of DAGs
• Probabilistic Aisubfield60%
  The paper is fundamentally about causal discovery, which is a core topic in probabilistic AI. It deals with structural causal models (SCMs), interventions, and conditional independence, all of which are central to probabilistic graphical models and causal inference.
  Structural Causal Models (SCMs), popularized by the works of Pearl [1998, 2009], graphically describe networks of causal dependencies.Causal discovery is the task of recovering the underlying causal structure from data, typically in the form of a directed acyclic graph (DAG) or a representation of an equivalence class of DAGs

Abstract

Causal discovery algorithms learn a network that describes the causal dependencies among random variables. A common workflow involves first utilizing conditional independence properties on observational data to determine partially directed causal relationships, then applying interventions to orient the unknown causal directions. A critical assumption for the first step is faithfulness: a requirement that causally linked variables exhibit statistical dependence. Many natural systems include buffering and stabilizing pathways that cancel out to achieve systemic robustness. This cancellation of pathways violates faithfulness, leading causal discovery algorithms to incorrectly remove causal dependencies. In this paper, we argue that hard interventions contain information about the presence/absence of causal linkage that is overlooked in the first stage of structure discovery. We show that a mild assumption -- called intervention-immediacy faithfulness -- that allows cancellations, is sufficient to nonparametrically identify causal structures with hard interventions. These results position interventions as the primary carriers of information about causal structure, which should take precedence over conditional independence testing. To flip the paradigm, we also specify equivalence classes when the identification criteria are not met due to limitations in the scope of interventions.

Paper Context