E9 203: Compressive Sensing and Sparse Signal Processing: Projects

Logistics

  • Projects must be done individually.
  • Select a pair among the software packages listed below, and send me an email about your selection. Once you have selected the pair, you must champion one, and your name will appear next to the package in [square brackets]. The second package should be a “different” algorithm. For example, if the first package is an l1-based recovery method, the second should be a non-convex or a greedy recovery method. Once a package is taken, other students will have to select a different package for their project.
  • Your grade will depend on:
  1. Your demonstration of your understanding of the package (through the presentation, report, simulations, etc)
  2. The presentation (organization, clarity, content)
  3. The report (details of the package, simulation results, critical comments)
  4. Any new ideas/results you can add, however small.
  • Please use this style file for writing your reports.
  • Initial project presentation (18 mins per group, 3 mins for questions and changing between speakers): Mar. 19th and 21st, 2014. For your presentation:
  1. First slide: Title of your project (paper), along with its complete reference, and your name, department, and email address.
  2. Briefly discuss the package that you are championing: Show the solution to HW1 using your package
  3. Present the main algorithm that the package uses. In your opinion, what is are the strengths/weaknesses of this approach?
  4. Make a proposal on what you will show in the final presentation and report
  5. Any other items you wish to include – you have 18 mins!
  • Initial project presentation schedule:
  1. Mar. 19th, 2014: TBD.
  2. Mar. 21st, 2014: TBD.
  • Final project presentation and report: you must aim to answer questions like (the more important ones are in bold):
  1. Classes of problems the package can solve
  2. Ability to handle noisy measurements, non-sparse signals
  3. Precise details of the algorithms implemented in the package
  4. How to set the values of run-time parameters of the algorithm, if any?
  5. Recovery guarantees for the algorithm (theoretical guarantees that can be given)
  6. Phase transition diagrams
  7. A detailed comparative report on the two software packages
  8. Complexity, numerical stability, and robustness
  9. Real-time implementability (e.g., can it be parallelized?)

Packages

  1. Stephen Becker’s CoSAMP and OMP code, paper Shubhanshu Shekhar – 2
  2. FOCUSS, k-t FOCUSS, tM FOCUSS
  3. GAMP
  4. L1-LS, user’s guide
  5. l1 magic Vikas Kumar Dewangan – 2
  6. SpaRSA Srikanth Raj – 2
  7. Yall1, user’s guide
  8. Zhilin_MSBL Vidhyadhar Upadhya – 1
  9. Sparse Bayesian Learning, see also David Wipf’s code Vidhyadhar Upadhya – 1
  10. L1-homotopy algorithms
  11. SPG L1
  12. NESTA Shubhanshu Shekhar – 1
  13. FPC AS
  14. WaveLab Vikas Kumar Dewangan – 1
  15. GPSR Srikanth Raj – 1
  16. SPAMS
  17. Stephen Becker’s CoSAMP and OMP code
  18. SPARCO
  19. SparseLab
  20. SLEP

Please email me if you find any other package that ought to be included.