Saturday, January 12, 2013

All hopes may not be crushed all at once

Here is some feedback from yesterday's It's not a bad reconstruction, just the end of an illusion...  Maybe this is a good way to restore one's confidence that somewhere, somehow there is a metric you cannot not like :-)

Hey Igor,
You mentioned in today's post that what you assume about the unknowns is pretty important. Another aspect that my student Jin Tan (copied) has been studying is what you're prioritizing in your reconstruction. Traditionally, signal processing has emphasized the square error. But Jin has work that can minimize the ell_1 error, the support set error,... - her algorithm is very flexible in supporting a broad range of possible error metrics, and this might come in handy in some applications.
Dror





No comments:

Printfriendly