Institute of Information Science, Academia Sinica



Press Ctrl+P to print from browser


Identifying Bad Measurements Via Separation in Compressive Sensing

  • LecturerMr. Tsung-Han Lin (Ph.D. candidate in Computer Science of Harvard University)
    Host: Dr. Jan-Ming Ho
  • Time2011-01-12 (Wed.) 14:00 – 16:00
  • LocationAuditorium 106 at new IIS Building



We consider the problem of identifying bad measurements in 
compressive sensing. These bad measurements can be present due to 
malicious attacks and system malfunction.
Since the linear system of equations in compressive sensing is 
underconstrained, errors introduced by these bad measurements can 
result in large changes in computed solutions. We describe a 
separation-based method for identifying bad measurements. In this 
method we separate out top non-zero variables by ranking, eliminate 
these variables from the system of equations, and then solve the 
reduced overconstrained problem to identify bad measurements. 
Comparing to prior methods based on direct or joint l1-minimization, 
the separation-based method can addresses challenging cases when 
there are only a moderate number of samples. In developing the method 
we introduce the notion of inversions which govern the separability 
of the non-zero variables.





Tsung-Han Lin is a Ph.D. candidate in Computer Science of Harvard 
University. He has received his B.S. and M.S. degree in the Department 
of Electrical Engineering, National Taiwan University. His research 
interests include wireless networks, compressive sensing, wireless 
networking for unmanned aerial vehicles (UAVs), and wireless sensor