Causal Inference with Attention (CInA) is a theoretically sound method that leverages multiple unlabeled datasets to perform self-supervised causal learning, enabling zero-shot causal inference on unseen tasks with new data.
본 연구는 널리 사용되는 K-평균 군집화 알고리즘을 활용하여 실험군 간 이질적 치료 효과의 잠재적 하위 집단 구조를 발견하는 새로운 방법론을 제안한다.
관찰 데이터를 활용하여 각 노드의 개별 처리 효과(ITE)를 추정하고, 이를 기반으로 감염된 노드들의 ITE 합을 최대화하는 영향력 최대화 알고리즘을 제안한다.
관찰 데이터에서 핵심 혼란 변수를 정확히 식별하고 이를 활용하여 편향 없는 인과 효과 추정을 제공하는 일반적인 인과 추론 프레임워크를 제안한다.
The key confounding covariates that require adjustment for unbiased causal inference in cross-sectional observational data are the common root ancestors of the treatment and outcome variables.
Neyman's repeated sampling framework can be used to experimentally evaluate the empirical performance of individualized treatment rules (ITRs), including those derived from modern causal machine learning algorithms, under a minimal set of assumptions.
This work introduces the problem of Causally Abstracted Multi-armed Bandits (CAMABs), where decision-making problems are modeled at different levels of resolution and related via causal abstraction. The authors propose algorithms to learn across these related models and analyze their regret.
A novel collaborative inverse propensity score weighting estimator that outperforms meta-analysis methods when dealing with heterogeneous data across multiple sites.
대규모 언어 모델은 상관관계 정보만으로 인과관계를 추론하는 데 어려움을 겪는다.
This article introduces a novel doubly robust estimator for average treatment effects in the presence of unobserved confounding, leveraging matrix completion techniques to handle large-scale data with repeated measurements across units.