Conformal prediction and causal inference
Ahmed Alaa, Professor
Electrical Engineering and Computer Science
Check back for status
Conformal prediction (CP) is a model-agnostic and distribution-free method for quantifying uncertainty in black-box machine learning (ML) models. CP can be used to construct prediction sets/intervals that covers the true labels with a pre-determined probability as long as the training and testing data are exchangeable. While this assumption may hold in a supervised learning setup, it does not hold in causal inference problems where the goal is to predict causal effects of an intervention on individual units. This project will explore the theory and methods for applying CP to a various causal inference problems.
Role: - Develop new theory and methods for CP in causal inference settings, and will be supervised directly by the PI.
- Students will be expected to meet with their supervisor at least once a week.
- Students will conduct literature reviews, develop , develop new algorithms and run experiments.
- Experience with Python is required.
Successful applicants should have a strong background in statistics, mathematics or theoretical computer science. The workload for this project is expected to be 12 hours/week or more.
Qualifications: - Solid foundation in statistics, mathematics, or theoretical computer science
- Completion of Stat 241B / CS 281B is highly desirable
- Strong interest in pursuing graduate studies
Day-to-day supervisor for this project: Lars van der Laan, Graduate Student
Hours: 12 or more hours
Related website: https://proceedings.neurips.cc/paper_files/paper/2023/hash/94ab02a30b0e4a692a42ccd0b4c55399-Abstract-Conference.html
Related website: https://alaalab.berkeley.edu/