Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DODO: Causal Structure Learning with Budgeted Interventions

Matteo Gregorini, Chiara Boldrini, Lorenzo Valerio

TL;DR

DODO tackles the problem of learning causal structure when the true DAG is hidden and interventions are costly. It combines observation, targeted interventions, and a pruning step based on partial correlations to identify a direct-edge causal graph within a linear Gaussian SCM. Empirical results show DODO outperforms observational baselines such as PC and NOTEARS across varying graph sizes and budgets, achieving near-perfect F1 in favorable settings and a notable +0.25 F1 gain in challenging cases. This approach offers a data-efficient path for autonomous agents to build accurate causal models under resource constraints, with implications for explainability and safe intervention planning.

Abstract

Artificial Intelligence has achieved remarkable advancements in recent years, yet much of its progress relies on identifying increasingly complex correlations. Enabling causality awareness in AI has the potential to enhance its performance by enabling a deeper understanding of the underlying mechanisms of the environment. In this paper, we introduce DODO, an algorithm defining how an Agent can autonomously learn the causal structure of its environment through repeated interventions. We assume a scenario where an Agent interacts with a world governed by a causal Directed Acyclic Graph (DAG), which dictates the system's dynamics but remains hidden from the Agent. The Agent's task is to accurately infer the causal DAG, even in the presence of noise. To achieve this, the Agent performs interventions, leveraging causal inference techniques to analyze the statistical significance of observed changes. Results show better performance for DODO, compared to observational approaches, in all but the most limited resource conditions. DODO is often able to reconstruct with as low as zero errors the structure of the causal graph. In the most challenging configuration, DODO outperforms the best baseline by +0.25 F1 points.

DODO: Causal Structure Learning with Budgeted Interventions

TL;DR

DODO tackles the problem of learning causal structure when the true DAG is hidden and interventions are costly. It combines observation, targeted interventions, and a pruning step based on partial correlations to identify a direct-edge causal graph within a linear Gaussian SCM. Empirical results show DODO outperforms observational baselines such as PC and NOTEARS across varying graph sizes and budgets, achieving near-perfect F1 in favorable settings and a notable +0.25 F1 gain in challenging cases. This approach offers a data-efficient path for autonomous agents to build accurate causal models under resource constraints, with implications for explainability and safe intervention planning.

Abstract

Artificial Intelligence has achieved remarkable advancements in recent years, yet much of its progress relies on identifying increasingly complex correlations. Enabling causality awareness in AI has the potential to enhance its performance by enabling a deeper understanding of the underlying mechanisms of the environment. In this paper, we introduce DODO, an algorithm defining how an Agent can autonomously learn the causal structure of its environment through repeated interventions. We assume a scenario where an Agent interacts with a world governed by a causal Directed Acyclic Graph (DAG), which dictates the system's dynamics but remains hidden from the Agent. The Agent's task is to accurately infer the causal DAG, even in the presence of noise. To achieve this, the Agent performs interventions, leveraging causal inference techniques to analyze the statistical significance of observed changes. Results show better performance for DODO, compared to observational approaches, in all but the most limited resource conditions. DODO is often able to reconstruct with as low as zero errors the structure of the causal graph. In the most challenging configuration, DODO outperforms the best baseline by +0.25 F1 points.

Paper Structure

This paper contains 24 sections, 11 equations, 5 figures, 1 table.

Table of Contents

  1. Introduction
  2. Background and Related Work
  3. Learning causality via observational data
  4. Learning causality through interventions
  5. Graph Sizes and Structures in Causal Discovery
  6. System model
  7. World model
  8. Agent model
  9. The DODO algorithm
  10. Observation Phase
  11. Intervention Phase
  12. Causal Links Detection
  13. Indirect Causal Connections Pruning
  14. Budget and Epochs
  15. Experimental settings
  16. ...and 9 more sections

Figures (5)

  • Figure 1: Observational Algorithms F1 score comparison with 95% confidence interval
  • Figure 2: F1 score for 20-node graphs; mean $\pm$ 95% confidence interval.
  • Figure 3: Structural Hamming Distance for 20-node graphs; mean $\pm$ 95% confidence interval.
  • Figure 4: F1 score for 10-node graphs; mean $\pm$ 95% confidence interval.
  • Figure 5: Structural Hamming Distance for 10-node graphs; mean $\pm$ 95% confidence interval.