Time and Room:

  • Time: Spring Quarter 2018, WF, 1:00 PM - 2:20 PM
  • Room: CMU 228

Instructor: Joseph Salmon
Emails: or

Course Web Page:

Office Hours:

  • Friday 10:30-11:30 AM, by appointment only.


Stat 538, and Stat 535 (or CSE 546) or an equivalent rst course in statistical learning.
Students are expected to be familiar with probability theory, multivariate analysis, linear
algebra, advanced calculus, and convex optimization. Students should be able to complete
short programming tasks using high level programming language (e.g. Python, Matlab, R/S-

Course description
This course will introduce the basis on robust statistics. On top of modeling and theoret-
ical aspects (in uence function, breaking point, depth, sensitivity curves, etc.), the course
will cover some numerical optimization for implementing the introduced methods. Time
permitting, each registered student will report on a topic of interest to her/him.

This course will be project oriented. Final projects will be proposed based on a selection of
a short list of articles or according to students' relevant propositions. This will be a credit
/ no credit course.

Topics include (as time permits):
1. Introduction. Examples Basic concepts, equivariance, breaking point,
2. Location/scale estimates, M-estimates, Pseudo-observations, depth
3. L-statistics: Linear combination of order statistics
4. G^ateaux dierentiability, Sensitivity curve, In uence Function
5. Numerical computation of M-estimates, reminders on non-smooth convex optimization,
Iterative Reweighted Least Square (IRLS)
6. Smoothing non smooth problems
7. Robust regression for multivariate statistics
8. Quantile regression, "crossing"