Project:	Evaluating Causal Machine Learning Algorithms
Disciplines:	Computer Science, Applied and Computational Mathematics
Mentor:	Schulman Leonard, Professor, (EAS), schulman@caltech.edu, Phone: 626 395 6839
Mentor URL:	https://users.cms.caltech.edu/~schulman/  (opens in new window)
AO Contact:	Erik Jahn, ejahn@caltech.edu
Background:	Distinguishing causation from correlation is a fundamental problem in all scientific disciplines. The framework of Graphical Causal Models (also called Causal Bayesian Networks) introduced by Judea Pearl provides a principled mathematical theory for tackling this problem. Many causal discovery algorithms have been developed for learning causal graphs from observational data based on different mathematical approaches such as conditional independence testing, score-based optimization, or sparse regression. We want to compare the practical performance of these approaches on various test datasets.
Description:	The student will implement multiple causal discovery algorithms (such as PC, GES, NOTEARS, CAM, LiNGAM) and evaluate their performance across different data sets and evaluation metrics. Moreover the student will devise some non-causal comparison methods such as simple correlation analysis or random guessing to provide performance baselines.
References:	Introduction to causal graphs: https://towardsdatascience.com/using-causal-graphs-to-answer-causal-questions-5fd1dd82fa90/ Survey on causal discovery methods: https://arxiv.org/pdf/2305.10032 Recent paper on the importance of random baselines: https://arxiv.org/abs/2412.10039
Student Requirements:	Coding experience with Python and a strong mathematical background is required.
Programs:	This AO can be done under the following programs:   Program    Available To        SURF    both Caltech and non-Caltech students  Click on a program name for program info and application requirements.