Causal Machine Learning using Tractable Circuits
Abstract
This project will pursue research directions at the intersection of causal reasoning and machine learning. It particularly aims to i,ntegrate observational and experimental data with causal knowledge, for the purpose of equipping learned systems with the ability to, perform interventional and counterfactual reasoning which are viewed as the marks of causal reasoning. Our approach will be based o,n the theory of tractable Boolean and arithmetic circuits which was pioneered by the PI, while exploiting some recent advances by th,e PI which show that even abstract forms of causal knowledge can significantly facilitate computation. We will also investigate a ne,w class of objective functions, called causal objective functions, and a novel optimization problem, called all-unit selection, whos,e goal is to find all objects that optimize a user-specified causal objective function. The ultimate goal of the project is to enhan,ce the scalability of causal reasoning through compilation into tractable circuits and to democratize such reasoning through the sys,tematic employment of causal objective functions. The project will realize these goals through a set of planned theoretical developm,ents and the construction of a corresponding software system to be released publicly. This abstract is approved for public release.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jul 13, 2022
- Source ID
- N000142212501
Entities
People
- Adnan Darwiche
Organizations
- Office of Naval Research
- United States Navy
- University of California, Los Angeles