P. Weiss : Concentration inequalities for the synthesis problem in compressive sampling
In this talk, I will present algorithms in order to solve the synthesis
problem in compressive sampling: for fixed representation and
measurement bases, how to design sampling patterns that guarantee
accurate reconstruction of sparse signals?
This work is based on a recent contribution of Juditski et al. who
proposed randomized algorithms that consist in selecting measurements
independently at random. We propose a simple convergence proof
based on concentration inequalities. This proof extends to more complex
frameworks where the sampling schemes are generated using random walks
on the space of measurements. Such sampling strategies are relevant for
practical imaging systems such as MRI where successive samples should
be close.
|