Alessandro Perelli
Mon 29 May 2017, 13:00 - 14:00
Classroom 7, Hudson Beare Building, King’s Buildings, EH9 3JL

If you have a question about this talk, please contact: Jonathan Mason (s1015431)

Image for MultiD-AMP: match up Accuracy and Fast Computation by Dynamically Denoising Data

Pizza will be at the second floor foyer at 12:30 for people who want to attend the seminar

ABSTRACT    
Denoising-Approximate Message Passing (D-AMP) refers to a class of iterative algorithms for image/signal reconstruction where at each iteration a non-linear denoising function is applied to the signal estimate.
In this work we aim at designing a mechanism for leveraging an hierarchy of denoising models (MultiD-AMP) in order to minimize the overall complexity given the expected risk, i.e. the estimation error, since the denoising is often the computational bottleneck in the D-AMP reconstruction.
The intuition comes from the observation that at earlier iteration, when the estimate is far according to some distance to the true signal, the algorithm does not need a complicated denoiser, since the structure of the signal is poor, but rather faster denoisers; this leads to the idea of defining an hierarchy of denoisers of increased complexity.  
The main challenge is to define a switching scheme which is based upon the empirical finding that in MultiD-AMP we can predict exactly, in the large system limit, the evolution of the Mean Square Error, based on a set of training images.
The proposed framework has been effectively tested on i.i.d. random Gaussian measurements with Gaussian noise and for fan beam X-ray Computed Tomography reconstruction.


BIOGRAPHY
Alessandro Perelli received the Bachelor of Science and Master of Science in Electronic Engineering respectively in 2007 and 2010 from the Università Politecnica delle Marche, Italy and the Ph.D. in Electronic Engineering from the University of Bologna, Italy, in 2014. He has been a visiting research scholar at the Ultrasound Group of University of Leeds, UK, from 2012 to 2013.
Since July 2014, he is a Research Associate at the Institute of Digital Communications (IDCOM) University of Edinburgh.
His current activities focus on reconstruction techniques for computed tomography, compressive Sensing, stochastic optimization and variational inference.