Publications

A new technique for causal inference that can sometimes learn more than a randomized controlled trial.

Domain expertise bias creates paradoxes in causality and covariate shift methods. We show these are the same as paradoxes in social choice theory.

The first known algorithm for causal discovery under latent class counfounding.

A (near) resolution of the sample complexity gap for mixtures of products.

We focus on a challenging distribution-shift setting in which the causal and anticausal variables of the target are unobserved. Leaning on information theory, we develop feature selection and engineering techniques for the observed downstream variables that act as proxies. We identify model-stabilizing proxies and moreover utilize auxiliary training tasks to answer counterfactual questions that stabilize our models.

The first known algorithm for $k$-MixBND.

Mutation profiles contain a signal for Glioblastoma prediction.

We develop the “method of synthetic bits” for solving discretem mixtures of product distributions, giving a exponential time complexity improvement (in the number of sources). The algorithm involves a reduction to the $k$-MixIID case.

Nontransitivite properties of the voting systems are also possible in networks of experts and machine learning models.

Importance weighting via probability density function estimation is unstable and requires parameter tuning. We demonstrate that reweighting according to cumulative distirbution functions is stable without parameter tuning.

We introduce mutation profiles as a way to capture time-dependent information in the repeat region of a genome and demonstrate the ability of mutation profiles to predict cancer-type from non-tumor DNA.

A patent for mutation profiles and their ability to predict disease propensity.

We use Prony’s method to solve the $k$-MixIID problem, which gives improved sample and time complexity via a new stability analysis.

We demonstrate that amplification processes lead to biases in TCGA data.