Research not for publishing papers, but for fun, for satisfying curiosity, and for revealing the truth.

This blog reports latest progresses in
(1) Signal Processing and Machine Learning for Biomedicine, Neuroimaging, Wearable Healthcare, and Smart-Home
(2) Sparse Signal Recovery and Compressed Sensing of Signals by Exploiting Spatiotemporal Structures
(3) My Works


Sunday, April 8, 2012

Presentation: COMPRESSED SENSING AND SPARSE SIGNAL RECOVERY BY SPARSE BAYESIAN LEARNING: MODELS, ALGORITHMS, AND APPLICATIONS

In this Thursday's Research Expo 2012 (April 12), I will present my previous and on-going work on sparse signal recovery/compressed sensing using sparse Bayesian learning (SBL).

Here is the abstract:

Compressed sensing / sparse signal recovery is a hot field in signal processing. Numerous algorithms have been proposed and have shown promising successes in applications. Among these algorithms, sparse Bayesian learning (SBL) has outstanding performance. In this presentation I will summarize our lab's recent work on SBL. I will present four new models that data-adaptively learn and exploit signals' temporal, spatial, spatiotemporal, and dynamic information. The derived algorithms from these models have shown the best, or at least top-tier, performance among existing compressed sensing algorithms in both computer simulations and practical applications (e.g. telemonitoring, biomarker selection in gene expression, source localization, earthquake detection, neuroimaging). Particularly, some of them have:

(1) solved the challenge of non-sparse physiological signals (e.g. fetal ECG contaminated by strong noise, EEG, EMG, etc) telemonitoring via wireless body-area networks with ultra-low power consumption, which was not solved before;  (this application will show SBL's ability to recover non-sparse signals with little distortion, using simple sparse binary matrices as its sensing matrices)
(2) achieved higher EEG source localization accuracy than other famous algorithms in more complicated environments (this application will show SBL's excellent ability under strong noise ( 0dB), highly coherent sensing matrix, and frequently changing signals);
(3) broke the record of predicting cognitive outcomes from neuroimaging measures in Alzheimer's disease in 2011 (showing SBL's ability to  deal with highly coherent sensing matrix);
(4) Brain-Computer Interface (solving the speed problem of SBL)
(5) obtained the best accuracy in earthquake detection in some common datasets;

Some of these work have been published or submitted to: IEEE Trans. on Signal Processing, IEEE Journal of Selected Topics in Signal Processing, Proceedings of the IEEE, IEEE Trans. on Biomedical Engineering, NeuroImage, CVPR 2012, and ICASSP 2010, 2011, 2012. Also, a US patent is pending.

Welcome to hear my presentation on the recent progress on various spatiotemporal SBL models!

------------------------------------------------------------------------
Photo: N. talangensis (taken by April 7)

No comments:

Post a Comment