PulseExploreJournal ClubDebatesTrendingResearchersJournals
Instagram
HomeExploreJournal ClubTrending
Synapse
⌘+K
Synapse
May 19, 20241 citationsOpen Access

Adaptive Online Experimental Design for Causal Discovery

View Full Paper
MEMuhammad Qasim ElahiLWLai WeiMKMurat Kocaoglu

Key Points

  • The proposed adaptive algorithm achieves higher accuracy in causal graph inference, utilizing fewer samples than traditional methods.
  • Achieving superior performance in simulations, the algorithm reduces the structural hamming distance between the learned graph and the true causal graph.
  • Online learning principles, inspired by bandit problems, drive the intervention selection process for efficient data utilization in causal discovery tasks across various settings and scenarios.

Abstract

Causal discovery aims to uncover cause-and-effect relationships encoded in causal graphs by leveraging observational, interventional data, or their combination. The majority of existing causal discovery methods are developed assuming infinite interventional data. We focus on data interventional efficiency and formalize causal discovery from the perspective of online learning, inspired by pure exploration in bandit problems. A graph separating system, consisting of interventions that cut every edge of the graph at least once, is sufficient for learning causal graphs when infinite interventional data is available, even in the worst case. We propose a track-and-stop causal discovery algorithm that adaptively selects interventions from the graph separating system via allocation matching and learns the causal graph based on sampling history. Given any desired confidence value, the algorithm determines a termination condition and runs until it is met. We analyze the algorithm to establish a problem-dependent upper bound on the expected number of required interventional samples. Our proposed algorithm outperforms existing methods in simulations across various randomly generated causal graphs. It achieves higher accuracy, measured by the structural hamming distance (SHD) between the learned causal graph and the ground truth, with significantly fewer samples.

Ask AI
Helpful
Bookmark
Share
View Full Paper

Cite This Study

Elahi et al. (2024) studied this question.

synapsesocial.com/papers/68e69709b6db64358761d67ahttps://doi.org/10.48550/arxiv.2405.11548
Ask AI
Helpful
Bookmark
Share
View Full Paper