Course Catalog 2007-2008

SGN-2756 ROBUST ESTIMATION, 3 cr
Robust Estimation

Courses persons responsible
Sari Peltonen

Lecturers
Sari Peltonen

Objectives
After passing this course the student understands what robustness means and knows which tools can be used for studying robustness of estimators (filters). Student will also become familiar with several robust estimators.

Content
Content Core content Complementary knowledge Specialist knowledge
1. The meaning of robustness, deviations from parametric models and estimation theory.  Chebychev's inequality, information inequality, unbiasedness, consistency, efficiency and basics of detection.  Minimax approach. 
2. Influence function (IF), gross-error sensitivity, local-shift sensitivity, rejection point, asymptotic variance, breakdown point.  Finite-sample versions of the IF and output distributional influence function (ODIF).    
3. Order Statistics (OS) and stack filters and their distributions.  Optimization of the stack filters.    
4. Definitions of M-, L-, and R-estimators.  Spefific properties of these different estimator classes. Redescending M-estimators and matched median filter.    

Requirements for completing the course
One seminar presentation and final exam.

Evaluation criteria for the course

  • Grading is pass/fail. In order to pass the student has to give seminar presentation and get at least half of the maximum points from the final exam.

  • Used assessment scale is passed / failed

    • Study material
    Type Name Auhor ISBN URL Edition, availability... Exam material Language
    Book Robust Statistics Hampel, Ronchetti, Rousseeuw & Stahel     John Wiley, 1985 Yes  English 
    Book Robust Statistics Huber     John Wiley, 1981 No  English 
    Lecture slides "Robust Estimation" Sari Peltonen   http://www.cs.tut.fi/kurssit/SGN-2756/   Yes  English 
    Book Robust Statistics Maronna Ricardo, Martin Doug and Yohai Victor 978-0-470-01092-1 http://www.wiley.com/legacy/wileychi/robust_statistics/ John Wiley, 2006 No  English 

    Prerequisites
    Code Course Credits M/R
    SGN-2706 SGN-2706 Nonlinear Signal Processing 5 Recommendable

    Prequisite relations (Sign up to TUT Intranet required)

    Remarks

    From the book by Hampel et al. the first two chapters are studied on this course.

  • Partial passing of course must be in connection with the same round of implementation.

  • The course is suitable for postgraduate studies.

  • Course will not be lectured in the academic year 2007-2008.

    • Distance learning

  • ITC utilized during the course

    • - In information distribution via homepage, newsgroups or mailing lists, e.g. current issues, timetables
    - In compiling teaching material, particularly for online use or other electronic media
    - In distributing and/or returning exercise work, material etc

  • Estimate as a percentage of the implementation of the course

    • - Contact teaching: 50 %
    - Distance learning: 0 %
    - Proportion of a student's independent study: 50 %

    Scaling
    Methods of instructionHours
    Lectures 30
    Seminar reports 48

    Other scaledHours
    Exam/midterm exam 3
    Total sum 81

    Principles and starting points related to the instruction and learning of the course

  • The first half of the course consists of lectures given for a group of maximum 20 students. On the second half of the course all the students give a seminar presentation about some related topic. Students are encouraged to ask questions both during/after the lectures and presentations.

    • Additional information related to course
    The course is lectured every other year.

    Correspondence of content
    8001552 Robust Estimation

    Course homepage

    Last modified 18.04.2007
    Modified bySari Peltonen