Harsh Parikh, PhD

"Credence uses state-of-the-art deep generative models such as variational auto-encoders (VAEs) to approximate the universe of complex datasets. These generative models are trained and validated on a collection of observed data sets. Credence uses these trained deep generative models to generate data that has analogous complexity to the observed data. Credence’s procedure enables users to have perfect knowledge about ground truth treatment effects of the intervention in the generated data. This allows the users to evaluate their method in a principled fashion without compromising on the complexity or the realness of the data they are evaluating the method on.

Credence learns a generative model by anchoring the level of endogeneity or treatment effect or anchoring both simultaneously. Anchoring the treatment effect and/or endogeneity is analogous to constraining the search space of potential data generators. Our approach can be conceptualized as projecting the true data-generative process to a constrained space of data-generators and finding the closest data-generator that conserves the joint distribution of X,Y,Z as close as possible to that of the observed data under the constraints."

Faculty: Harsh Parikh, PhD

Download: GitHub / credence-v2 package

Platform: Python

Reference: arxiv.org (credence-v2)

---

We introduce a flexible framework that produces high-quality almost-exact matches for causal inference. Most prior work in matching uses ad-hoc distance metrics, often leading to poor quality matches, particularly when there are irrelevant covariates. In this work, we learn an interpretable distance metric for matching, which leads to substantially higher quality matches. The learned distance metric stretches the covariate space according to each covariate’s contribution to outcome prediction: this stretching means that mismatches on important covariates carry a larger penalty than mismatches on irrelevant covariates. Our ability to learn flexible distance metrics leads to matches that are interpretable and useful for the estimation of conditional average treatment effects.

Faculty: Harsh Parikh, PhD

Download: GitHub / MALTS package

Platform: Python, R

Reference: jmlr.org (MALTS)

---

Randomized controlled trials (RCTs) serve as the cornerstone for understanding causal effects, yet extending inferences to target populations presents challenges due to effect heterogeneity and underrepresentation. Our paper addresses the critical issue of identifying and characterizing underrepresented subgroups in RCTs, proposing a novel framework for refining target populations to improve generalizability. We introduce an optimization-based approach, Rashomon Set of Optimal Trees (ROOT), to characterize underrepresented groups. ROOT optimizes the target subpopulation distribution by minimizing the variance of the target average treatment effect estimate, ensuring more precise treatment effect estimations. Notably, ROOT generates interpretable characteristics of the underrepresented population, aiding researchers in effective communication.

Faculty: Harsh Parikh, PhD

Download: GitHub / ROOT package

Platform: Python, R

Reference: arxiv.org (ROOT)

---