causal-inference

FedCausal-Dyn is a novel federated learning framework that addresses dynamic feature drift by separating causal features from spurious variations, enabling robust prototype aggregation. It achieves state-of-the-art performance on federated domain generalization benchmarks.

This paper proposes causal workloads—differentially private query sets based on orthogonal moments—to enable valid causal inference from synthetic data, introducing methods like Causal-AIM and noise-aware multiple imputation.

Introduces CausalDS, a benchmark for evaluating causal reasoning in LLM-based data science agents, using synthetic structural causal models and natural language stories to test associational, interventional, and counterfactual reasoning along with tool use and abstention.

This paper theoretically and empirically studies the relationship between counterfactual fairness (CF) and group fairness (GF) in image classification, introducing new CF evaluation datasets (CelebA-CF and LFW-CF). It finds that CF does not imply GF in images due to latent attributes correlated with sensitive attributes, and proposes Counterfactual Knowledge Distillation (CKD) to mitigate this.

This paper introduces Heckman-corrected epistemic uncertainty to address selection on unobservables in machine learning, demonstrating that importance weighting fails when selection depends on unobservables correlated with outcomes. The method restores calibration in controlled experiments and real data, outperforming standard UQ baselines.

This paper introduces a human-in-the-loop framework for personalized algorithmic recourse that iteratively approximates a user's causal model through Bayesian inference, improving the plausibility and cost-effectiveness of recommendations.

A study analyzing 142K NLP papers from 2010–2026 finds that both established and new NLP authors are increasingly publishing in general ML venues like NeurIPS and ICLR rather than core NLP conferences like ACL, with a significant citation premium favoring ML venues.

This paper uses AT&T's exclusive iPhone monopoly from 2007-2011 as a natural experiment to estimate the causal effect of iPhone ownership on birth rates, finding evidence that the iPhone may act as a form of birth control.

This paper proposes a causal auditing framework to evaluate forgetting in Limited Memory Language Models by varying the database state during inference, discovering that parametric leakage is negligible and post-deletion correctness primarily arises from retrieval artifacts rather than residual parametric memory.

This paper from Airbnb combines economic modeling and causal inference to understand how guests respond to prices and how preferences vary, aiming to optimize pricing tools and personalization in the two-sided marketplace.

该论文提出贝叶斯上下文实验者（Bayesian in-context experimenters），通过训练Transformer模仿贝叶斯后验Neyman教师策略，实现自适应平均处理效应（ATE）估计，并采用混合专家Transformer处理未知平滑性，理论证明可通过监督预训练学习该策略。

This paper presents a causal machine learning approach combining double/debiased machine learning with a hierarchical Bayesian framework to estimate the incremental impact of additional supply on marketplace outcomes, using Airbnb as a case study.

This paper presents an integrated framework for personalizing free-value thresholds in a two-sided job marketplace, addressing competing objectives and constrained experiments. The deployed system shows significant lift in target metrics while respecting engagement guardrails.

CausalMix formulates data mixture optimization as a causal inference problem for LLM training, enabling dynamic adaptation to shifting data distributions without costly retraining, and demonstrates improved performance on Qwen2.5-0.5B and Qwen3-4B-Base.

This article recommends the top 10 skills and tools for social science research, including Auto-Empirical-Research-Skills developed by the Stanford team, for using AI agents to conduct empirical research and write papers.

This paper introduces lifted causal inference, leveraging parametric causal factor graphs to efficiently compute causal effects in relational domains, and presents the Lifted Causal Inference (LCI) algorithm for polynomial-time inference.

TabPFN-CFM is a causal foundation model that predicts both causal structure and outcomes from observational data, supporting all three levels of Pearl's Causal Hierarchy and achieving improved performance over baselines.

This paper conducts a same-hands re-evaluation of bivariate causal direction methods on the Tübingen cause-effect pairs, introducing a parameter-free compression baseline that ties with SLOPE. It documents how published accuracy figures are inflated by protocol differences and releases all code and data.

This survey provides a systematic review of federated causal discovery and inference, organizing methods by methodological paradigm, federation topology, and structural scope, and highlighting open challenges.

This paper introduces causal reinforcement learning (CRL), unifying causal inference and reinforcement learning under a structural causal model framework, and explores novel learning settings such as generalized policy learning and counterfactual learning.