FEO3310 Sparse Signal Processing

Instructor: Saikat Chatterjee

---

News: The course begins 24th Feb, 2014, 13:00-15:00; Class room: SIP Seminar Room, Plan 3, Q-building, Osquldas Väg 10. Call 0738913581 if door is closed.

---

The course is focused on solving a sparse solution of a linear under-determined system with the trade-off between complexity and performance. Most relevant applications are compressed sensing and low rank matrix reconstruction. A brief outline of the course contents is as follows.

• The key problem of solving a linear under-determined system of equations and sparsity
• Pursuit algorithms - Design and their theoretical performance guarantees
• From exact to approximate solutions
• Iterative-shrinkage algorithms
• Towards average performance analysis
• The Dantzig-Selector algorithm
• MAP versus MMSE estimation 
• RIP based analysis
• Random sampling in bounded orthonormal systems

General information

Course credit: 8 points 
Instructor: Saikat Chatterjee 
Instructor's office: Osquldas väg 10, floor 4. 
Instructor's email: sach@kth.se 
Meeting times: For discussion, mail to Instructor and fix a time 
Class Room: The list will be given in the 'Schedule' table below.
Work load: 2 hours per lecture and assignment. 
Prerequisites: Linear algebra and elementary convex optimization. 
Teaching and learning methodology: The lectures will be based on blackboard and slides. Students have to present papers and execute projects.

Evaluation

The evaluation criteria are the presentation of research papers and project assignments. Each paper presentation will be graded according to (approximate thresholds): 
 -1 : less that 20% of the paper is understood correctly 
  0 : between 20 % to 40% of the paper is understood correctly 
  1 : between 40% to 70% of the paper is understood correctly 
  2 : more than 70% of the paper is understood correctly 
There are three paper presentations (two in a group as technical notes preparations and one individually as a paper presentation) and the threshold for receiving a pass-grade is four points out of six points. For the group work of technical note preparations, a group will be evaluated as a whole. For individual presentation, the person will be treated individually.

In addition, the student has to successfully complete two project assignments. The project assignments will mainly focus on implementing algorithms and their use in practice. Each project assignment will be graded according to (approximate thresholds): 
 -1 : less that 20% of the project executed 
  0 : between 20 % to 40% of the project executed 
  1 : between 40% to 70% of the project executed 
  2 : more than 70% of the project executed 
The threshold for receiving a pass-grade is three points out of four points.

Overall, for achieving a pass-grade, the threshold is seven points out of ten points. If required, a final examination of five hours may be arranged.

Course textbooks

• Michael Elad, “Sparse and Redundant Representations: From Theory to Applications in Signal and Image Precessing” 2010, Springer
• Simon Foucart and Holger Rauhut, "A Mathematical Introduction to Compressive Sensing'', 2013, Birkhäuser

Schedule and lecture notes

• Course class lecture span: 11th Feb 2014 to 13th March 2014.
• Total ten classes, over five weeks.
• All classes are of two hours, from 13:00-15:00. 

Details

List of papers for presentation

For paper presentation, each group consists of 2 persons. Each group has to read two good papers and make a technical note followed by presentation. Please find some papers which you may choose.

According to interest, you may choose other quality papers. Please inform the paper title to the instructor.

Objective: For writing good technical notes, the group members should do intense discussion and exchange of ideas, revolving around the chosen papers. By this way, students are expected to learn from each other.

Tasks for technical note writing: (1) Read the paper with most attention. (2) Understand the central issues - what is the research and why and how? (3) If theorems are there, try to relate with reality. (4) You may not need to rework all the proofs exactly - but, for your own learning, try to go as far as possible. (5) Then write a good summary of the research work (within 2-3 pages). (6) Present the work so that audience understands the summary. (7) Provide a thought process such that the papers can be useful for your own research.

Presentation dates: Two dates will be decided later. If we do not finish within these two days, then we may finish by another later date. I prefer that the presentations should be finished by these dates. However, if somebody wants more time, that can be arranged.

Presentation time limit: Approximately one hour. Ultimately, it depends on the audience interactions.

List of projects

For projects, we are interested in hands on learning. The first project will be given by the instructor to the group. The second project is individual where a student has to choose a good paper (mostly dealing with a practical problem) and understand the paper followed by reproducing the results in the paper. The individual paper work has to be presented which will be counted as the individual paper presentation as well as individual project presentation. The overall format is as follows.

Project submission and presentation date: Will be decided later. It is preferable that the presentations should be finished by this date. However, if somebody wants more time, that can be arranged.

Presentation time limit: Approximately 40 minutes. Ultimately, it depends on the audience interactions.

The details of the group project assigned by the istructor is here. Download from here.

Ultimate Objective

Being a new research area with lots of activities, the purpose of the course will be served if the students are able come up with new research results.