STATS364: Selective Inference

Syllabus

Instructor & TAs

Instructors

Jonathan Taylor

• Email

• Office hours: M 3:30-5:30

Email list

The course has an email list: stats364-spr1920-staff@lists.stanford.edu

Schedule

MW 1:30-2:50, Zoom

Prerequisites

A course in inference of 300A.

Course description

This course is focused on inferential methods in the context where, as Candes has put it: “Scientists collect data first and ask questions later.” This practice necessitates the need for inferential tools with statistical guarantees recognizing these questions are derived from the data we hope to use to answer such questions with hypothesis tests, confidence intervals, posterior distributions, etc.

Common approaches to this problem include simultaneous approaches (e.g. multiple comparisons procedures that control FWER or FDR) and conditional (e.g. adaptive clinical trials such as “drop the losers”).

This class will have the format of a reading group mixed with some lectures. We may have some guest lecturers TBD.

Generally speaking, Mondays will be a lecture and Wednesdays a discussion of assigned reading.

Rough outline

Below is a rough outline of our first 6 weeks. Subject to changes in pace, interest in topics among students in the class.

Week	Monday lecture	Wednesday reading
1 (M: 4/6)	Course introduction (notebook)	Benjamini, Y. Simultaneous and selective inference: Current successes and future challenges. Taylor, J. and Tibshirani, R. Statistical learning and selective inference Brief notes
2 (M: 4/13)	Multiple comparisons I (notebook)	Berk et al. Valid Post-Selection Inference Taylor, J. A selective survey of selective inference Brief notes
3 (M: 4/20)	Multiple comparisons II Guest lecturer: Arun Khuchibhotla Supplement	Lockhart et al. + Discussants A significance test for the LASSO Zhong and Prentice Bias-reduced estimators and confidence intervals for odds ratios in genome-wide association studies. Brief notes
4 (M: 4/27)	Conditional inference I (simple problem) (notebook)	Drop the losers (notebook) Leeb and Potscher Can one estimate the conditional distribution of post-model-selection estimators? Lee, Sun, Sun and Taylor Exact post-selection inference, with application to the lasso Brief notes
5 (M: 5/4)	Conditional inference II (LASSO)	Fithian, Sun and Taylor Optimal inference after selection Brief notes
6 (M: 5/11)	Conditional inference III (General setup) Data carving (notebook)	Fithian, Taylor, Tibshirani and Tibshirani Selective sequential model selection (Discussed in discussion of Fithian, Sun, Taylor) Candes and Barber Controlling the false discovery rate via knockoffs (Delayed in schedule)
7 (M: 5/18)	Knockoffs for multiple comparisons	Lei and Fithian AdaPT: An interactive procedure for multiple testing with side information (Delayed in schedule) Li and Foygel Barber Accumulation tests for FDR control in ordered hypothesis testing (Delayed in schedule)
8	Memorial Day	Conditional inference IV (CLT) Asymptotics (notebook)
9 (M: 6/1)	Approximate likelihood (notebook)	Black box (notebook) Weinstein and Ramdas Online Control of the False Coverage Rate and False Sign Rate (Delayed in schedule)
10	Review (notebook)	Projects

Evaluation

• 3-4 shortish assignments (50%)

• project (50%)

Notes

• Introduction HTML, notebook

[HTML]

Assignments