CS: What's up ? Are we in the 100% or the 0.01% improvement business ? Two examples: Turbo-AMP and an illumination camera.

Nearly a year ago, I was asking if we were hitting a ceiling for reconstruction solvers. Well it looks we have a serious contender in Turbo-AMP as presented by Phil Schniter in his recent presentation entitled: Turbo-AMP: A Graphical-Models Approach to Compressive Inference

 

The only one I really know is IRWL1 which is pretty slow in the first place. So a tenfold improvement over a slow solver is probably nothing much to brag about but AMP, as the presentation shows, has the ability to add additional assumptions in the signal being sought besides sparsity. So if speed is not the main discriminator, it looks like being able to make more assumptions on the signal is the way to go. For pure kick, I'd love to see a comparison between Turbo-AMP and GPSR or SPGL1.  Obviously, we would not be having this conversation if something along the lines of an automatic benchmarking tool existed in the ethers, uh ? Anyway, this is a very interesting development in an area thought to have flat lined. I recall being excited about AMP but this new development (turbo-AMP) feels good.

In a totally different direction, I am not quite sure what to make of this paper: An Active Illumination Single-Pixel Camera Based on Compressive Sensing by Filipe Magalhaes, Francisco Araujo, Miguel Correia, Mehrdad Abolbashari, and Faramarz Farahi. The abstract reads:

In this paper an optical imaging system based on compressive sensing (CS) is presented along with its principal mathematical aspects. Although CS is undergoing significant advances and empowering many discussions and applications throughout various fields, this article focus on the analysis of a single-pixel camera. This work was the core for the development of a single-pixel camera approach based on active illumination. Therefore, the active illumination concept is described along with experimental results, which were very encouraging towards the development of compressive sensing based cameras for various applications, such as pixel level programmable gain imaging.

I guess this is the first publication of an illumination camera but I thought the Arizona folks had already done that. I recall that while the Rice folks featured it at a conference, they did not seem to have a publication on that. Anyway, I'll add it to the compressive sensing hardware page.