Yesterday, I jumped to conclusions in saying that the LB-101M chip was performing the algorithm mentioned in the presentation of Stephane Mallat at ECTV'08. After asking him specifically that question, he answered back stating the LB-101M chip did not implement the algorithm of the presentation as I initially stated. He also stated that the chip implemented a somewhat smarter algorithm given the hardware constraints. In the description one can see these constraints, the chip "was implemented on a small Altera ™ Cyclone-II 70 FPGA, with only 70000 logic elements." If you have ever worked with an FPGA, you know how much space a multiplication takes, and one can only be amazed at what this chip is doing.
The NIPS '08 online papers are out. From the Rice Compressive Sensing Repository, there is a new paper I have not covered before:
Justin Haldar, Diego Hernando, and Zhi-Pei Liang, Compressed sensing in MRI with non-Fourier encoding. The abstract reads:
Compressed sensing (CS) has inspired significant interest because of its potential to reduce data acquisition time. There are two fundamental tenets to CS theory: (1) signals must be sparse or compressible in a known basis, and (2) the measurement scheme must satisfy specific mathematical properties (e.g., restricted isometry or incoherence properties) with respect to this basis. While MR images are often compressible with respect to several bases, the second requirement is only weakly satisfied with respect to the commonly used Fourier encoding scheme. This paper explores the possibility of improved CS-MRI performance using non-Fourier encoding, which is achieved with tailored spatially-selective RF pulses. Simulation and experimental results show that non-Fourier encoding can significantly reduce the number of samples required for accurate reconstruction, though at some expense of noise sensitivity. Theoretical guarantees and their limitations in MRI applications are also discussed.
Rachel Ward renamed "Cross validation in compressed sensing via the Johnson-Lindenstrauss Lemma" into Compressed sensing with cross validation.