Don’t worry about people stealing your ideas. If it’s original, you’ll have to ram it down their throats.
I think this portrays well some of the conundrum we in the compressed sensing community, face when it comes to a narrative about the usefulness of compressed sensing. The narrative developed in Emmanuel Candes talk is a good one, but to most of you reading this piece, it's already an old story and it's certainly not one that is going to make you a rock star (unless you are part of the group of people featured there). But then you are going to ask, where are the low hanging fruits, where should I go next ? I was recently talking to somebody about a technology that has been using many of the solvers devised initially for compressive sensing. The narrative in that field seems to call everything that has an l_1 solver a compressive sensing problem. The most breathtaking aspect of this example is that the acquisition systems have remained the same, i.e.coherent while at the same time most reconstruction using l_1 solvers have been doing moderately better than traditional solvers. And so, the consensus from that community is that compressive sensing offers no real improvement over current results. Well duh!.... This community has simply surrendered to what I call Lena's syndrome, an all to common straw man argument: Use a degraded benchmark that replaces synthetically the acquisition process and then show that reconstruction of the signal cannot be better than with domain specific solution techniques developed over the years. Don't get me wrong, I am sure people are going to get tenure over this, that some people's egos are going to go up because they have a paper on the latest shiny thing or that we are going to read some special edition of journals showing the compressive sensing may bring a 10% improvement blah blah blah... But did we get in this business to play in this theater ? I can see that y'all are shaking your heads. To be more specific, one of the argument mentioned in that community is that the measurement matrix is not fulfilling RIP, not random enough, blah blah blah, you know the sort of argument your kids tell you when you're serving them Broccolis, oh it's not green enough, it's not steamed enough. Yeah that's right, these Broccolis are not the sweet chocolate bars you've been yearning for but you really have not other option, ... really. Well if the measurement matrix is not random enough, why aren't you spending all your waking hours making it random ? you'll realize then that maybe you will have to change the whole hardware because of that, and then you'll realize that even if you are not solving your initial problem, you may be solving a more profound one or many others at once. You signed up for that, right ?