Streaming Covariate Balancing via Discrepancy-Based Feature Coresets

Real-time estimation of average treatment effects (ATE) in streaming observational data poses two key challenges: strict memory constraints that preclude storing the full data history, and distributional shifts in both treatment assignment and outcome-generating process. Existing methods either require offline access to the entire dataset for covariate balancing or rely on parametric online models that are vulnerable to model misspecification under such shifts. This paper proposes a novel model-agnostic method for ATE estimation in streaming data, which effectively addresses the above challenges. Based on discrepancy theory, we first compress streaming data into feature coresets that preserve covariate balancing objectives over a rich nonparametric function class, enabling linear-time updates with bounded memory. Then, by directly learning balancing weights and bypassing parametric propensity score estimation, we enhance the model's robustness against the shift in treatment assignment, while by balancing over an expressive function space we make the model more adaptive to the shift in the outcome-generating process. Theoretically, we establish convergence guarantees with explicit bounds on memory usage and computational complexity. Empirically, extensive experiments on both synthetic and real-world datasets show the effectiveness and robustness of the proposed method, consistently outperforming existing techniques.

Many applications need to estimate the effect of a treatment or action from data that arrives continuously, such as medical records or online user data. This is challenging because we may not be able to store all past data, and the data patterns can change over time. We propose a new method that keeps only a small but informative summary of the data and uses it to make fair comparisons between treated and untreated groups. The method avoids strong modeling assumptions, making it more reliable when treatment behavior or outcomes shift. We prove that our method is accurate while using limited memory and computation. Experiments on simulated and real-world data show that it is more robust than existing approaches.