Representation Learning as a Tool for Causal Discovery
Abstract
Massive data collection holds the promise of a better understanding of complex phenomena and ultimately, of better decisions. Repres,entation learning has become a key driver of machine learning applications, since it allows learning latent spaces that capture impo,rtant properties of the data without requiring any supervised annotations. While representation learning has been hugely successful,in predictive tasks, it can fail miserably in causal tasks including predicting the effect of an intervention. This calls for a marr,iage between representation learning and causal inference. An exciting opportunity in this regard stems from the growing availabilit,y of interventional data (in medicine, advertisement, education, etc.). However, these datasets are still miniscule compared to the,action spaces of interest in these applications (e.g., interventions can take on continuous values like the dose of a drug or can be, combinatorial as in combinatorial drug therapies). We will develop a framework for using representation learning (in particular inf,initely wide autoencoders that have been shown to perform well in applications and can be analyzed mathematically since they corresp,ond to neural tangent kernels and can be computed in closed-form) as a tool for causal discovery to deal with exploding action space,s. By learning causal representations, we will enable predicting the effect of unseen interventions across different contexts (synth,etic interventions for causal imputation) as well as identify the optimal interventions to move a system to a desired state (optimal, intervention design). While the developed methods will be broadly applicable, the methods will be validated in the application to d,rug screening as well as cellular reprogramming, where large-scale interventional datasets are publicly available.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jan 14, 2022
- Source ID
- N000142212116
Entities
People
- Caroline Uhler
Organizations
- Massachusetts Institute of Technology
- Office of Naval Research
- United States Navy