Math @ Duke

Publications [#258028] of David B. Dunson
search www.stat.duke.edu.Papers Published
 Ji, S; Dunson, D; Carin, L, Multitask compressive sensing,
Ieee Transactions on Signal Processing, vol. 57 no. 1
(2009),
pp. 92106, ISSN 1053587X [doi]
(last updated on 2018/08/19)
Abstract: Compressive sensing (CS) is a framework whereby one performs N nonadaptive measurements to constitute a vector v∈ℝN with v used to recover an approximation u∈RℝM to a desired signal u∈RℝM with N≪ M; this is performed under the assumption that uis sparse in the basis represented by the matrix Ψ∈RℝM×M. It has been demonstrated that with appropriate design of the compressive measurements used to define v, the decompressive mapping v⇁ umay be performed with error ∥uu∥22 having asymptotic properties analogous to those of the best transformcoding algorithm applied in the basis Ψ. The mapping v⇁u constitutes an inverse problem, often solved using ℓ1 regularization or related techniques. In most previous research, if L〉 sets of compressive measurements vii=1,L are performed, each of the associated uii=1,L are recovered one at a time, independently. In many applications the "tasks"defined by the mappings vi⇁ ui are not statistically independent, and it may be possible to improve the performance of the inversion if statistical interrelationships are exploited. In this paper, we address this problem within a multitask learning setting, wherein the mapping vi ⇁uifor each task corresponds to inferring the parameters (here, wavelet coefficients) associated with the desired signal ui, and a shared prior is placed across all of the L tasks. Under this hierarchical Bayesian modeling, data from all L tasks contribute toward inferring a posterior on the hyperparameters, and once the shared prior is thereby inferred, the data from each of the L individual tasks is then employed to estimate the taskdependent wavelet coefficients. An empirical Bayesian procedure for the estimation of hyperparameters is considered; two fast inference algorithms extending the relevance vector machine (RVM) are developed. Example results on several data sets demonstrate the effectiveness and robustness of the proposed algorithms. © 2008 IEEE.


dept@math.duke.edu
ph: 919.660.2800
fax: 919.660.2821
 
Mathematics Department
Duke University, Box 90320
Durham, NC 277080320

