Perfect Match: A Simple Method for Learning Representations For Counterfactual Inference With Neural Networks (2018-09-27T00:00:00.000000Z)

TL;DR

Perfect Match is presented, a method for training neural networks for counterfactual inference that is easy to implement, compatible with any architecture, does not add computational complexity or hyperparameters, and extends to any number of treatments.

Abstract

Learning representations for counterfactual inference from observational data is of high practical relevance for many domains, such as healthcare, public policy and economics. Counterfactual inference enables one to answer "What if...?" questions, such as "What would be the outcome if we gave this patient treatment $t_1$?". However, current methods for training neural networks for counterfactual inference on observational data are either overly complex, limited to settings with only two available treatments, or both. Here, we present Perfect Match (PM), a method for training neural networks for counterfactual inference that is easy to implement, compatible with any architecture, does not add computational complexity or hyperparameters, and extends to any number of treatments. PM is based on the idea of augmenting samples within a minibatch with their propensity-matched nearest neighbours. Our experiments demonstrate that PM outperforms a number of more complex state-of-the-art methods in inferring counterfactual outcomes across several benchmarks, particularly in settings with many treatments.

Authors

P. Schwab

3 papers

Lorenz Linhardt

2 papers

W. Karlen

1 papers

TL;DR

Abstract

Authors

References44 items

Causality

Limits of Estimating Heterogeneous Treatment Effects: Guidelines for Practical Algorithm Design

Confounding-Robust Policy Improvement

A comparison of methods for model selection when estimating individual treatment effects

General-purpose validation and model selection when estimating individual treatment effects

GANITE: Estimation of Individualized Treatment Effects using Generative Adversarial Nets

Not to Cry Wolf: Distantly Supervised Multitask Learning in Critical Care

Granger-Causal Attentive Mixtures of Experts: Learning Important Features with Neural Networks

Elements of Causal Inference: Foundations and Learning Algorithms

Deep Counterfactual Networks with Propensity-Dropout

Causal Effect Inference with Deep Latent-Variable Models

Bayesian Inference of Individualized Treatment Effects using Multi-task Gaussian Processes

Generalized random forests

Recursive Partitioning for Personalization using Observational Data

Estimating individual treatment effect: generalization bounds and algorithms

Assessing the Gold Standard--Lessons from the History of RCTs.

Learning Representations for Counterfactual Inference

Importance Sampling for Minibatches

Estimation and Inference of Heterogeneous Treatment Effects using Random Forests

Reputation and Power: Organizational Image and Pharmaceutical Regulation at the FDA

bartMachine: Machine Learning with Bayesian Additive Regression Trees

The Cancer Genome Atlas Pan-Cancer analysis project

Approximate Nearest Neighbor: Towards Removing the Curse of Dimensionality

MatchIt: Nonparametric Preprocessing for Parametric Causal Inference

Doubly robust estimation of causal effects.

Scikit-learn: Machine Learning in Python

Bayesian Nonparametric Modeling for Causal Inference

Domain Adaptation: Learning Bounds and Algorithms

BART: Bayesian Additive Regression Trees

Matching as Nonparametric Preprocessing for Reducing Model Dependence in Parametric Causal Inference

Applied Bayesian Modeling and Causal Inference From Incomplete-Data Perspectives

The Propensity Score with Continuous Treatments

Propensity Score-Matching Methods for Nonexperimental Causal Studies

Does Matching Overcome Lalonde's Critique of Nonexperimental Estimators?

Identification and Estimation of Causal Effects of Multiple Treatments Under the Conditional Independence Assumption

The Role of the Propensity Score in Estimating Dose-Response Functions

Evaluating the Econometric Evaluations of Training Programs with Experimental Data

The central role of the propensity score in observational studies for causal effects

2018) between 0.1, 1, and 10. For the GANITE generators and discriminators, we used MLP architectures

Measuring living standards with proxy variables

Bayesian Additive Regression Trees

Causal inference using potential outcomes: Design, modeling, decisions

Random Forests

Computing Infrastructure used computing infrastructure consisting of compute nodes with Intel i5 and Xeon CPUs to run experiments described in this paper

Field of Study

Journal Information

Name

Volume

Venue Information

Name

Type

URL

Alternate Names