arXiv:2408.01582v1 Announce Kind: new
Summary: Estimating remedy results from observational knowledge is of central curiosity throughout quite a few utility domains. Particular person remedy impact presents probably the most granular measure of remedy impact on a person stage, and is probably the most helpful to facilitate personalised care. Nonetheless, its estimation and inference stay underdeveloped on account of a number of challenges. On this article, we suggest a novel conformal diffusion model-based method that addresses these intricate challenges. We combine the extremely versatile diffusion modeling, the model-free statistical inference paradigm of conformal inference, together with propensity rating and covariate native approximation that sort out distributional shifts. We unbiasedly estimate the distributions of potential outcomes for particular person remedy impact, assemble an informative confidence interval, and set up rigorous theoretical ensures. We reveal the aggressive efficiency of the proposed technique over present options via intensive numerical research.
Supply hyperlink