Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Test-Time Learning of Causal Structure from Interventional Data

Wei Chen, Rui Ding, Bojun Huang, Yang Zhang, Qiang Fu, Yuxuan Liang, Han Shi, Dongmei Zhang

TL;DR

This work designs a self-augmentation strategy to generate instance-specific training data at test time, effectively avoiding distribution shifts, and developed a PC-inspired two-phase supervised learning scheme, which effectively leverages self-augmented training data while ensuring theoretical identifiability.

Abstract

Supervised causal learning has shown promise in causal discovery, yet it often struggles with generalization across diverse interventional settings, particularly when intervention targets are unknown. To address this, we propose TICL (Test-time Interventional Causal Learning), a novel method that synergizes Test-Time Training with Joint Causal Inference. Specifically, we design a self-augmentation strategy to generate instance-specific training data at test time, effectively avoiding distribution shifts. Furthermore, by integrating joint causal inference, we developed a PC-inspired two-phase supervised learning scheme, which effectively leverages self-augmented training data while ensuring theoretical identifiability. Extensive experiments on bnlearn benchmarks demonstrate TICL's superiority in multiple aspects of causal discovery and intervention target detection.

Test-Time Learning of Causal Structure from Interventional Data

TL;DR

This work designs a self-augmentation strategy to generate instance-specific training data at test time, effectively avoiding distribution shifts, and developed a PC-inspired two-phase supervised learning scheme, which effectively leverages self-augmented training data while ensuring theoretical identifiability.

Abstract

Supervised causal learning has shown promise in causal discovery, yet it often struggles with generalization across diverse interventional settings, particularly when intervention targets are unknown. To address this, we propose TICL (Test-time Interventional Causal Learning), a novel method that synergizes Test-Time Training with Joint Causal Inference. Specifically, we design a self-augmentation strategy to generate instance-specific training data at test time, effectively avoiding distribution shifts. Furthermore, by integrating joint causal inference, we developed a PC-inspired two-phase supervised learning scheme, which effectively leverages self-augmented training data while ensuring theoretical identifiability. Extensive experiments on bnlearn benchmarks demonstrate TICL's superiority in multiple aspects of causal discovery and intervention target detection.
Paper Structure (62 sections, 5 theorems, 26 equations, 14 figures, 19 tables, 5 algorithms)

This paper contains 62 sections, 5 theorems, 26 equations, 14 figures, 19 tables, 5 algorithms.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Interventional Causal Discovery
  4. Joint Causal Inference Framework
  5. Identifiability with Intervention Data
  6. Test-Time Learning of Causal Structure
  7. Training Data Acquisition via Self-Augmentation
  8. Two-Phase Supervised Causal Learning
  9. Experiments
  10. Causal Structure Identification Performance (RQ1)
  11. Training Data Study: Quantity and Quality (RQ2)
  12. Sampling and Running Efficiency Study (RQ3)
  13. Different Intervention Settings Study (RQ4)
  14. Conclusion and Future work
  15. Background Knowledge and Related Work
  16. ...and 47 more sections

Key Result

Theorem 1

Assume that JCI Assumptions ass:simple_scm, ass:uncaused and ass:unconfounded hold for SCM $\mathcal{M}$: For any other SCM $\tilde{\mathcal{M}}$ satisfying JCI Assumptions ass:simple_scm, ass:uncaused and ass:unconfounded that is the same as $\mathcal{M}$ except that it models the context differently, i.e., of the form with $\mathcal{J} \subseteq \tilde{\mathcal{J}}$ and $\text{\normalfont\scsh

Figures (14)

  • Figure 1: (Empirical dominance across interventional SCL tasks). TICL consistently outperforms all SoTA methods on both $\mathcal{I}$-CPDAG discovery and intervention targets detection, across diverse intervention families and limited-sample regimes.
  • Figure 2: Left: The workflow of test-time learning of causal structure. Right: An identifiability example.
  • Figure 3: The overall workflow of TICL for test-time learning of causal structure from interventional data.
  • Figure 4: The impact of different synthetic data strategies on the performance of causal structure learning.
  • Figure 5: The impact of different initialization strategies on the performance of IS-MCMC convergence.
  • ...and 9 more figures

Theorems & Definitions (18)

  • Definition A.1: Directed Acyclic Graph
  • Definition A.2: Skeleton
  • Definition A.3: UT and V-structures
  • Definition A.4: $PC$
  • Definition A.5: Vicinity
  • Definition A.6: Sepsets
  • Definition A.7: Causal Graph
  • Definition A.8: Structural Causal Model
  • Theorem 1
  • Corollary 2
  • ...and 8 more