NeurIPS 2019 Workshop

“Do the right thing”: machine learning and causal inference for improved decision making

December 14, 2019, Vancouver, Vancouver Conventional Center, West Level 1, West Ballroom C

Workshop Summary

In recent years, machine learning has seen important advances in its theoretical and practical domains, with some of the most significant applications in online marketing and commerce, personalized medicine, and data-driven policy-making. This dramatic success has led to increased expectations for autonomous systems to make the right decision at the right target at the right time.  This gives rise to one of the major challenges of machine learning today that is the understanding of the cause-effect connection. Indeed, actions, intervention, and decisions have important consequences, and so, in seeking to make the best decision, one must understand the process of identifying causality. By embracing causal reasoning autonomous systems will be able to answer counterfactual questions, such as “What if I had treated a patient differently?”, and “What if had ranked a list differently?” thus helping to establish the evidence base for important decision-making processes. 

The purpose of this workshop is to bring together experts from different fields to discuss the relationships between machine learning and causal inference and to discuss and highlight the formalization and algorithmization of causality toward achieving human-level machine intelligence.

Invited Speakers

Susan Athey

Stanford University

Susan Murphy

Harvard University

Andrea Rotnitzky

Harvard University and Universidad di Tella

Susan Shortreed

Kaiser Permanente Washington Health Research Institute

Ying-Qi Zhao

University of Washington and the Fred Hutchinson Cancer Research Center

Program Schedule

}

08:45am – 09:00am

West Ballroom C

Opening Remarks

By Nathan Kallus

}

09:00am – 09:30am

West Ballroom C

Confidence Intervals for Policy Evaluation in Adaptive Experiments

By Susan Athey

Invited speaker
Adaptive experiments can result in considerable cost savings in multi-armed trials by enabling analysts to quickly focus on the most promising alternatives. Most existing work on adaptive experiments (which include multi-armed bandits) has focused maximizing the speed at which the analyst can identify the optimal arm and/or minimizing the number of draws from sub-optimal arms. In many scientific settings, however, it is not only of interest to identify the optimal arm, but also to perform a statistical analysis of the data collected from the experiment. Naive approaches to statistical inference with adaptive inference fail because many commonly used statistics (such as sample means or inverse propensity weighting) do not have an asymptotically Gaussian limiting distribution centered on the estimate, and so confidence intervals constructed from these statistics do not have correct coverage. But, as shown in this paper, carefully designed data-adaptive weighting schemes can be used to restore a relevant central limit theorem, enabling hypothesis testing.
}

09:30am – 10:00am

West Ballroom C

Optimal adjustment sets in non-parametric graphical models

By Andrea Rotnitzky

Invited speaker
We consider the selection potential confounding variables at the stage of the design of a planned observational study. With such aim, given a tentative non-parametric graphical causal model the goal is to select the set of covariates that both suffices to control for confounding under the model and is satisfactory according to some pre-determined criterion. In this talk we focus on the criterion according to which an adjustment set is preferable to another if it yields a non-parametric estimator of some causal contrast of interest with smaller asymptotic variance. For studies aimed at assessing the effect of point exposure, static or dynamic, trx regimes we derive two graphical criteria: one to compare certain pairs of adjustment sets and a second to determine the optimal adjustment set. We show that these graphical rules coincide with rules derived in a recent article by Henckel et al, 2019, assuming linear causal graphical models and treatment effects estimated via ordinary least squares. For point exposure static regimes we also provide a rule for determining the optimal adjustment set among minimal adjustment sets. For studies aimed at assessing the effects of interventions at multiple time points, static or dynamic, we derive a graphical rule for comparing certain pairs of time dependent adjustment sets but we show that no global graphical rule is possible for determining optimal time dependent adjustment sets.
}

10:00am – 10:15am

West Ballroom C

Poster Spotlights

Two minute poster presentation
* Steve Yadlowsky, Hongseok Namkoong, Sanjay Basu, John Duchi, Lu Tian – Bounds on the Conditional and Average Treatment Effect with Unobserved Confounding Factors

* Marie-Laure Charpignon, Elizabeth Manrao, Min Kang, Kirti Kamboj, Kamal Choudhary, John Hoegger – Moving beyond correlation: Using a causal inference approach to measure the impact of a customer’s attribute on product retention A case study on Microsoft Teams collaborative app

* Maja Rudolph – Multi-Cause Inference with Sequential Treatments

* Amanda Coston, Alexandra Chouldechova, Edward Kennedy – Counterfactual Risk Assessments, Evaluation, and Fairness

* Yuta Saito – Unbiased Pairwise Learning from Implicit Feedback

* Jeremy Yang , Dean Eckles , Paramveer Dhillon, Sinan Aral – Optimizing Targeting Policies via Sequential Experimentation for User Retention

* Alexander Markham, Moritz Grosse-Wentrup – Measurement Dependence Inducing Latent Causal Models

}

10:15am – 11:00am

West Ballroom C

Coffee break, posters, and 1-on-1 discussions

Poster Reminders
– There are no poster boards at workshops. Posters are taped to the wall with the special tabs that the NeurIPS staff needs to order.
– Please make your posters 36W x 48H inches or 90 x 122 cm.
– Posters should be on lightweight paper, not laminated.
}

11:00am – 11:30am

West Ballroom C

Challenges in Developing Learning Algorithms to Personalize Treatment in Real Time

By Susan Murphy

Invited speaker
A formidable challenge in designing sequential treatments is to determine when and in which context it is best to deliver treatments. Consider mobile health behavioral interventions for individuals struggling with chronic health conditions. Operationally designing the sequential treatments involves the construction of decision rules that input current context of an individual and output a recommended treatment. There is much interest in personalizing the decision rules, particularly in real time as the individual experiences sequences of treatment. Here we discuss our work in designing and implementing an online reinforcement learning algorithm for use in improving physical activity among individuals with stage 1 hypertension.
}

11:30am – 12:00pm

West Ballroom C

Estimation and inference on high dimensional individualized treatment rule using split-and-pooled de-correlated score

By Ying-Qi Zhao

Invited speaker
Individualized treatment rules aim to identify if, when, which, and to whom treatment should be applied. One untackled question is the inference of the optimal individualized treatment rule. We propose a procedure for the simultaneous estimation and inference of such treatment rule with the existence of high dimensional covariates.The estimation procedure estimates the optimal individualized treatment rule as a weighted classification problem, while enjoying double robustness property. The inference procedure utilizes the data splitting, data pooling, and the semiparametric de-correlated score to conquer the slow convergence rate of estimated outcome regression or propensity score. The asymptotic properties for this procedure and its extensions are investigated via an analysis combining the techniques in high dimensional data and the doubly robustness. Simulation and real data analysis are conducted to justify the superiority of the proposed estimation and inference procedure under various settings.
}

12:00pm – 12:45pm

West Ballroom C

How machine learning, and causal inference work together: cross-pollination and new challenges

By David Sontag, Susan Athey, Susan Murphy, Andrea Rotnitzky, Ying-Qi Zhao, Susan Shortreed

Panel discussion
}

12:45pm – 02:30pm

West Ballroom C

Lunch

}

02:30pm – 03:00pm

West Ballroom C

Variable selection for causal inference: outcome-adaptive lasso

By Susan Shortreed

Invited speaker
The outcome-adaptive lasso is a variable selection approach for causal inference in observational settings. Traditionally, a “throw in the kitchen sink” approach has been used to select covariates for inclusion into the propensity score, but recent work shows including unnecessary covariates can impact both the bias and statistical efficiency of propensity score estimators. The outcome-adaptive lasso selects covariates for inclusion in propensity score models to account for confounding bias while maintaining statistical efficiency. This approach can perform variable selection in the presence of a large number of spurious covariates, that is, covariates unrelated to outcome or exposure. We will illustrate covariate selection using the outcome-adaptive lasso, including comparison to alternative approaches, using simulated data and in a survey of patients using opioid therapy to manage chronic pain.
}

03:00pm – 03:30pm

West Ballroom C

Oral Spotlights

Five minute oral presentation
* Daniel L Chen – Deep IV in Law: Analysis of Appellate Impacts on Sentencing Using High-Dimensional Instrumental Variables

* Jörn Boehnke, Pietro Bonaldi – Synthetic Regression Discontinuity: Estimating Treatment Effects using Machine Learning

* Yixin Wang, Dhanya Sridhar, David Blei – Equal Opportunity and Affirmative Action via Counterfactual Predictions

* Divyat Mahajan, Amit Sharma – Preserving Causal Constraints in Counterfactual Explanations for Machine Learning Classifiers

* Khashayar Khosravi, Vasilis Syrgkanis, Greg Lewis – Non-Parametric Inference Adaptive to Intrinsic Dimension

* Alexander D’Amour, Alexander Franks – Covariate Reduction for Weighted Causal Effect Estimation with Deconfounding Scores

}

03:30pm – 03:45pm

West Ballroom C

Poster Spotlights

Two minute poster presentation
* Théophile Griveau-Billion, Ben Calderhead – Efficient structure learning with automatic sparsity selection for causal graph processes

* Rahul Singh, Liyang Sun – De-biased Machine Learning for Compliers

* Zichen Zhang, Qingfeng Lan, Lei Ding, Yue Wang, Negar Hassanpour, Russ Greiner – Reducing Selection Bias in Counterfactual Reasoning for Individual Treatment Effects Estimation

* Jon Richens, Ciarán M. Lee, Saurabh Johri – Counterfactual diagnosis

* Jesse Krijthe, Tom Heskes – Estimating the long-term effect of early treatment initiation in Parkinson’s disease using observational data

* Bradley Butcher, Vincent S Huang, Jeremy Reffin, Sema K Sgaier, Grace Charles, Novi Quadrianto – Causal datasheet: An approximate guide to practically assess Bayesian networks in the real world

* Victor Chernozhukov, Whitney Newey, Vira Semenova – Inference on the weighted average value function with high-dimensional state space

* Rahul Ladhania – A Sequence of Two Studies to Propose & Test Sub-groups with Heterogeneous Treatment Effects

* Miruna Oprescu, Vasilis Syrgkanis, Keith Battocchi – EconML: A Machine Learning Library for Estimating Heterogeneous Treatment Effects

}

03:45pm – 04:30pm

West Ballroom C

Coffee break, posters, and 1-on-1 discussions

Poster Reminders
– There are no poster boards at workshops. Posters are taped to the wall with the special tabs that the NeurIPS staff needs to order.
– Please make your posters 36W x 48H inches or 90 x 122 cm.
– Posters should be on lightweight paper, not laminated.
}

04:30pm – 04:45pm

West Ballroom C

Closing Remarks

Call for Papers

We solicit submissions of novel research related to all aspects of causal inference, counterfactual prediction, and autonomous action. This includes, but is not limited to, the following topics:

  • Predicting counterfactual outcomes
  • Reinforcement Learning and Causal Inference
  • Causal transfer learning
  • Mediation analysis
  • Estimation of (conditional) average treatment effects
  • Contextual bandit algorithms and on-policy learning
  • Batch/offline learning from bandit feedback
  • Off-policy evaluation and learning
  • Interactive experimental control vs. counterfactual estimation from logged experiments
  • Online A/B-testing vs. offline A/B-testing
  • De-biasing observational data and feedback cycles
  • Fairness of actions and causal aspects of fairness
  • Applications in online systems (e.g. search, recommendation, ad placement)
  • Applications in physical systems (e.g. cars, smart homes)
  • Applications in medicine (e.g. personalized treatment, clinical trials)

We suggest extended abstracts of 2 pages in the NeurIPS format, but no specific format is enforced. The review process is double-blind. A maximum of 8 pages will be considered. References will not count towards the page limit. PDF files only. At the discretion of the organizers, accepted contributions will be assigned slots as contributed talks and others will be presented as posters.

Accepted Papers

Daniel L Chen – Deep IV in Law: Analysis of Appellate Impacts on Sentencing Using High-Dimensional Instrumental Variables

Jörn Boehnke, Pietro Bonaldi – Synthetic Regression Discontinuity: Estimating Treatment Effects using Machine Learning

Yixin Wang, Dhanya Sridhar, David Blei – Equal Opportunity and Affirmative Action via Counterfactual Predictions

Divyat Mahajan, Amit Sharma – Preserving Causal Constraints in Counterfactual Explanations for Machine Learning Classifiers

Khashayar Khosravi, Vasilis Syrgkanis, Greg Lewis – Non-Parametric Inference Adaptive to Intrinsic Dimension

Alexander D’Amour, Alexander Franks – Covariate Reduction for Weighted Causal Effect Estimation with Deconfounding Scores

Steve Yadlowsky, Hongseok Namkoong, Sanjay Basu, John Duchi, Lu Tian – Bounds on the Conditional and Average Treatment Effect with Unobserved Confounding Factors

Marie-Laure Charpignon, Elizabeth Manrao, Min Kang, Kirti Kamboj, Kamal Choudhary, John Hoegger – Moving beyond correlation: Using a causal inference approach to measure the impact of a customer’s attribute on product retention A case study on Microsoft Teams collaborative app

Maja Rudolph – Multi-Cause Inference with Sequential Treatments

Amanda Coston, Alexandra Chouldechova, Edward Kennedy – Counterfactual Risk Assessments, Evaluation, and Fairness

Yuta Saito – Unbiased Pairwise Learning from Implicit Feedback

Jeremy Yang , Dean Eckles , Paramveer Dhillon, Sinan Aral – Optimizing Targeting Policies via Sequential Experimentation for User Retention

Alexander Markham, Moritz Grosse-Wentrup – Measurement Dependence Inducing Latent Causal Models

Théophile Griveau-Billion, Ben Calderhead – Efficient structure learning with automatic sparsity selection for causal graph processes

Rahul Singh, Liyang Sun – De-biased Machine Learning for Compliers

Zichen Zhang, Qingfeng Lan, Lei Ding, Yue Wang, Negar Hassanpour, Russ Greiner – Reducing Selection Bias in Counterfactual Reasoning for Individual Treatment Effects Estimation

Jon Richens, Ciarán M. Lee, Saurabh Johri – Counterfactual diagnosis

Jesse Krijthe, Tom Heskes – Estimating the long-term effect of early treatment initiation in Parkinson’s disease using observational data

Bradley Butcher, Vincent S Huang, Jeremy Reffin, Sema K Sgaier, Grace Charles, Novi Quadrianto – Causal datasheet: An approximate guide to practically assess Bayesian networks in the real world

Victor Chernozhukov, Whitney Newey, Vira Semenova – Inference on the weighted average value function with high-dimensional state space

Rahul Ladhania – A Sequence of Two Studies to Propose & Test Sub-groups with Heterogeneous Treatment Effects

Miruna Oprescu, Vasilis Syrgkanis, Keith Battocchi – EconML: A Machine Learning Library for Estimating Heterogeneous Treatment Effects

Yangyi Lu, Amirhossein Meisami, Ambuj Tewari – Learning Good Interventions Sequentially via Causal Bandits

Julius von Kügelgen, Alexander Mey, Marco Loog, Bernhard Schölkopf – Semi-Supervised Learning, Causality and the Conditional Cluster Assumption

Niranjani Prasad – Defining Admissible Rewards for High-Confidence Policy Evaluation

David Simchi-Levi, Yunzong Xu, Jinglong Zhao – Unifying Adaptivity and Switching Ability in Online Experimental Design

Dmytro Mykhaylov, David J Rohde, Flavian Vasile, Martin Bompaire, Olivier Jeunen – Three Methods for Training on Bandit Feedback

Finnian Lattimore, David J Rohde – Causal inference with Bayes rule

Chirag Modi, Uros Seljak – Generative Learning of Counterfactual for Synthetic Control Applications in Econometrics

Negar Hassanpour, Russell Greiner – Learning Disentangled Representations for Counter Factual Regression

Ankit Sharma, Garima Gupta, Ranjitha Prasad, Arnab Chatterjee, Lovekesh Vig, Gautam Shroff – MetaCI: Meta-Learning for Causal Inference in a Heterogeneous Population

Yuta Saito, Gota Morishita – Dual Learning Algorithm for Delayed Feedback in Display Advertising

Shyngys Karimov, Gert Bijnens, Joep Konings – The impact of automatic wage indexation on Belgian employment, a machine learning approach

Jean Pouget-Abadie – Variance Reduction in Bipartite Experiments through Correlation Clustering

Jeroen Berrevoets, Sam Verboven, Wouter Verbeke – Optimising Individual-Treatment-Effect Using Bandits

Nan Rosemary Ke, Olexa Bilaniuk, Anirudh Goyal, Chris J Pal, Yoshua Bengio – Fast Adaptation and Slow Meta-Learning of Neural Causal Models

Peter Wirnsberger, Jovana Mitrovic, Melissa Tan, Charles Blundell, Lars Buesing – Improved Decision Making in Structural Causal Bandits

Julius von Kügelgen, Paul Rubenstein, Bernhard Schölkopf, Adrian Weller – Optimal experimental design via Bayesian optimization: active causal structure learning for Gaussian process networks

Logan Graham, Ciarán M. Lee, Yura Perov – Copy, paste, infer: A robust analysis of twin networks for counterfactual inference

Candice Schumann, Zhi T Lang, Nicholas Mattei, John P Dickerson – Group Fairness in Bandit Arm Selection

Anisha Zaveri, Victor Veitch – Sense and Sensitivity Analysis: Simple Post-Hoc Analysis of Bias Due to Unobserved Confounding

Anish Agarwal, Romain Cosson, Devavrat Shah, Dennis Shen – Bridging Randomized Control Trials and Personalized Treatments

Julie Josse, Imke Mayer, Jean-Philippe Vert – MissDeepCausal: causal inference from incomplete data using deep latent variable models

Victor Chernozhukov, Denis Nekipelov, Vira Semenova, Vasilis Syrgkanis – Regularized estimation of high-dimensional semiparametric models

Mert Demirer, Vasilis Syrgkanis, Greg Lewis, Victor Chernozhukov – Semi-Parametric Efficient Policy Learning with Continuous Actions

Vasilis Syrgkanis, Victor Lei, Miruna Oprescu, Maggie Hei, Keith Battocchi, Greg Lewis – Heterogeneous Treatment Effects with Instruments

Suraj Nair, Yuke Zhu, Silvio Savarese, Li Fei-Fei – Visual Causal Induction for Goal Directed Tasks

Aahlad Manas Puli, Rajesh Ranganath – Generalized Control Functions for Flexible IV Estimation

Dhanya Sridhar, Victor Veitch, David Blei – Using Text Embeddings for Causal Inference

Nathan Kallus, Masatoshi Uehara – Double Reinforcement Learning for Efficient Off-Policy Evaluation in Markov Decision Processes

Mohammadi Zaki, Avinash Mohan, Aditya Gopalan – Towards Optimal and Efficient Best Arm Identification in Linear Bandits

Issa Dahabreh – Causally interpretable meta-analysis with double/debiased machine learning

Yi Ding, Guillaume Basse, Panos Toulis – Minimax Crossover Designs

Ignavier Ng, Shengyu Zhu, Zhitang Chen, Zhuangyan Fang – A Graph Autoencoder Approach to Causal Structure Learning

Alan Mishler, Niccolo Dalmasso – When the Oracle Misleads: Modeling the Consequences of Using Observable Rather than Potential Outcomes in Risk Assessment Instruments

Eli Sherman, David Arbour, Ilya Shpitser – Policy Interventions Under Interference

Duncan Wadsworth – Measuring Fairness in A/B Testing

Shuxi Zeng, Pengchuan Zhang, Denis Charles, Eren Manavoglu, Emre Kiciman – Robust Neural Network for Causal Invariant Features Extraction

Aleksander Wieczorek, Volker Roth – Information Theoretic Causal Effect Quantification

Shuxi Zeng, Murat Bayir, Denis Charles, Joseph J Pfeiffer, Emre Kiciman – Causal Transfer Random Forest: Leveraging Observational and Randomization Studies

Hao Liu, Anqi Liu, Tongxin Li, Saeed Karimi, Yisong Yue, Animashree Anandkumar – Disentangling Causal Effects from Latent Confounders using Interventions

Khizar Qureshi, Tauhid Zaman – Running Influence Campaigns on Twitter

Kyra Gan, Andrew Li, Zachary Lipton, Sridhar Tayur – Causal Inference with Selectively-Deconfounded Data

Jason S Hartford, Kevin Leyton-Brown – Identifying Valid Instruments via Effect Agreement

Organizers

Thorsten Joachims

Cornell University

Nathan Kallus

Cornell University

Adith Swaminathan

Microsoft Research

Michele Santacatterina

Cornell University

David Sontag

MIT

Angela Zhou

Cornell University