The Sparsest Solution of Underdetermined Linear System by l q minimization for 0 < q 1

Size: px
Start display at page:

Download "The Sparsest Solution of Underdetermined Linear System by l q minimization for 0 < q 1"

Transcription

1 The Sparsest Solution of Underdetermined Linear System by l q minimization for 0 < q 1 Simon Foucart Department of Mathematics Vanderbilt University Nashville, TN Ming-Jun Lai Department of Mathematics, University of Georgia Athens, GA 3060 January 7, 008 Abstract We study tne l q approximation of the sparsest solution of underdetermined linear systems. Mainly we present a condition on the matrix associated with an underdetermined linear system under which the solution of l q minimization is the sparsest solution of the system. Our condition generalizes a similar condition in [Candés, Romberg and Tao 06] ensuring that the solution of the l 1 minimization is the sparsest solution. Our condition in l 1 case slightly improves the similar condition. We present a numerical method to compute the l q minimization. Our numerical experiments show that the l q method is better than many existing methods, to find the sparsest solution. 1 Introduction Given a matrix A of size m n with m < n, let R k = {Ax, x R n, x 0 k} be the range of A of all the k-component vectors, where x 0 denotes the number of nonzero entries in the vector x and k < m. Given a vector y R k, we are looking for a solution x R n with x 0 k such that y = Ax. In general, there may have many such solutions. The research problem is to find the sparsest one. That is, given y R k, solve the following minimization problem simon.foucart@vanderbilt.edu mjlai@math.uga.edu min{ x 0, x R n, Ax = y}. (1) 1

2 The solution of the above problem is called the sparsest solution of y = Ax. Although the problem in Eq. (1) is a NP problem (cf. [18]) in general, it can be solved by using many methods, e.g., l 1 minimization, OGA(orthogonal greedy algorithm) and BP(basis pursuit) under some assumptions such that the (P1) solution is also the (P0) solution. The l 1 minimization problem is the following min{ x 1, x R n, Ax = y}. () where x 1 = n i=1 x i for x = (x 1, x,, x n ) T. The solution 1 Ax is called the l 1 solution of y = Ax. There are several conditions on the matrix A that the l 1 solution is the sparsest solution (cf. [Donoho and Elad 03] [13]) and [Candes, Romberg, Tao 06][3]). One is to use the concept spark of matrix A. One is to use the so-called Restricted Isometry Property(RIP). Recall spark(a) is the smallest possible number σ such that there exists σ columns from A that are linearly dependent. It is clear that σ(a) rank(a) + 1. Theorem 1.1 ([Donoho and Elad 03]) A representation y = Ad is necessarily the sparsest possible if x 0 < spark(a)/. The concept of RIP can be explained as follows. Letting 0 < k < m be an integer and A T be a submatrix of A consisting of columns from A with column indices in T {1,,, n}, the k restricted isometry constant δ k of A is the smallest quantity such that (1 δ k ) x A Tx (1 + δ k) x for all subset T with #(T) k. If a matrix A has such a constant δ k > 0 for some k, A possesses RIP. With this concept, it is easy to see that if δ k < 1, then the solution of Eq. (1) is unique. Furthermore, Theorem 1. ([Candés, Romberg, and Tao 06]) Suppose that k 1 such that δ 3k + 3δ 4k < and let x R n be a vector with x 0 k. Then for y = Ax, the solution of Eq. () is unique and equal to x. This result is recently simplified slightly in the following way: Theorem 1.3 ([Candés 07]) Suppose that k 1 such that δ k < 1. Let x R n be a vector with x 0 k. Then for y = Ax, the solution of Eq. () is unique and equal to x.

3 In this paper, we study the following approach to approximate the (P 0 ) solution. Let x q = ( n x i q ) 1/q i=1 be the standard l q quasi-norm for 0 < q < 1. We consider the minimization for 0 < q < 1 min{ x q q, x Rn, Ax = y}. (3) A solution of the above minimization is denoted by q Ax. This is not a brand new approach. It was studied recently in [7]. It is shown that Theorem 1.4 ([Chartrand 07a]) Let q (0, 1]. Suppose that there exists a positive number b > 1 such that the matrix A has RIP such that δ ak + bδ (a+1)k < b 1 for a = b q/( q). Then the solution of Eq. (3) is the sparest solution. This above result was further simplified in [8] in the following sense. Corollary 1.1 ([Chartrand 07b]) Suppose that δ k+1 < 1. Then there is q > 0 such that for any y = Ax with x 0 k, the solution of l q minimization in Eq.(3) is unique and is x. However, the proof is incorrect. In his proof, a > 1 because of b and a < 1+1/k. Then no such a so that ak can be an integer. In this paper we study the l q minimization in Eq. (3) by generalizing and refining the proof in [Candes, Romberg, and Tao 06]. Comparing to the results above, our results are a slight improvement of Theorem 1. and Theorem 1.3 and can be derived a correct version of Corollary 1.1. For convenience, let us state our theorems here and leave the proof to a section later. To this end, we need the following notations: cond k (A) = max cond(a T ) = #(T) k max T {1,,n} #(T) k In addition, let α k, β k be the best constants such that largest eigenvalue of A T A T smallest eigenvalue of A T A. T α k z Az β z for any vector z R n with z 0 k. In fact, α k = 1 δ k and β k = 1 + δ k. Our main result in this paper is 3

4 Theorem 1.5 Fix q (0, 1]. Suppose that c l := cond l (A) < 1 + ( ) /q 1 l, k for some integer l > k. Then every k-sparse vector x is exactly recovered by q Ax, the minimization solution of Eq. (3). Moreover, one has for every x R n x q Ax q C(q, c l )σ k (x) q and x q Ax C(q, c l) k 1/q 1/σ k(x) q, (4) where σ k (x) q := min z 0 k x z q and C(q, c l ) is a positive constant. Two cases are of particular interest: one is when l = k and q = 1. That is, Corollary 1. If the condition cond k (A) < 3 (5) is satisfied, then the l 1 solution is the sparsest one and the estimates in Eq. (4) hold. In terms of the RIP, since cond k (A) = 1+δ k 1 δ k, the condition (5) is that δ k < 1/. This improves Theorem 1.. The other case of interest is when l = k +1 and all q > 0. In this case, we may assume cond k+ (A) < +. Indeed, there exists a q small enough such that ( ) 1/q 1/ k + 1 cond k+ (A) < 1 +. k It follows from Theorem 1.5 that Corollary 1.3 Suppose that cond k+ (A) < +, then the solution q (Ax) is exactly the x with x 0 k for q > 0 small enough and independent of x. It is easy to see that cond k+ (A) < + is equivalent to λ k+ < 1. The above corollary can be given in the formulation of Corollary 1.1. Corollary 1.4 Suppose that δ k+ < 1. Then there is q > 0 such that for any y = Ax with x 0 k, the solution of l q minimization in Eq. (3) is unique and is x. It is easy to see that our result is an extension of Theorem 1. from the l 1 minimization to l q minimization for 0 < q 1. Also we improve the result in Theorem 1.3 and correct Corollary 1.1. In addition, we also consider the sparsest solution from imperfect measurements. Under the assumption that the measurements y = Ax+e contain a noise vector e with e θ for a fixed amount θ > 0, we need to solve min{ x 0, x R n, Ax y θ}. (6) 4

5 We again approximate the sparsest solution using the l q minimization for 0 < q 1. For q = 1, the above problem was studied in [Candes, Romberg, and Tao 06]. In the same way, we generalize their results from q = 1 to a general q (0, 1]. min{ x 0, x R n, Ax y θ}. (7) The paper is organized as follows. We begin with discussing the existence of the solution of the above minimization. Then we give a proof of Theorem 1.5. Next we use a random matrix theory to show that there are matrices obey our condition for large values of S with overwhelming probability. Finally we derive a numerical algorithm to compute appproximate solution of Eq. (3) and Eq. (7). We numerically experiment our scheme and compare with the existing l 1 method based on the matlab programs available on Candes webpage, the newly available reweighted l 1 method, the orthogonal greedy algorithm ( OGA) (cf. [0]) and regularized optimial matching pursuit (ROMP) (cf. [xxx]). Numerical results show that our algorithm is better than the l 1 method, the reweighted l 1 method, and is also better than the orthogonal greedy method (OGA) (cf. [0]) and ROMP method. Proof of Main Results We first show that the minimization problem Eq. (3) has a solution for q > 0. That is, the existence of the solution is independent of the RIP and the spark of A. Theorem.1 Fix 0 < q < 1. There exists a solution q Ax solving Eq. (3). Proof. Without loss of generality, we may assume that A is of full rank and the first m columns of A are linearly independent. Then we can find a quick solution x 0 R n such that y = Ax 0 by setting all x i = 0 for m < i n. Let ν be the infinimum of Eq. (3). Then it is easy to see that 0 ν x 0 q q. Let x j R n be a sequence such that Ax j = y and x j q q ν, i.e., x j q q ν + 1/j as j +. Thus, writing x j = (x j 1,, xj n )T, for each i between 1 and n, we claim that x j i is bounded from the above by B = max{1, (ν + 1)1/q } < +. If x j i 1, we are done. Otherwise, we have x j i q ν + 1/j ν + 1. Therefore we only need to look for the minimizer over the bounded interval [ B, B] n. Since the minimization functional is continuous, it achieves its minimal over a bounded domain. Thus, the minimizer of Eq. (3) exists. 5

6 Next we study when the solution of the l q minimization i.e., the solution of Eq. (3) is the sparsest solution, the solution of Eq. (1). Mainly we now turn to the proof of Theorem 1.5. Proof of Theorem 1.5. We begin recalling the following properties for 0 < q < 1: x 1 x q, x q n 1/q 1/ x, x + y q q x q q + y q q for any x,y R n. Step 1: Null space property. We establish first that, for any index set S with S 0 k and any η ker(a), one has η S q ρ η S q, with ρ := condl (A) 1 ( ) 1/q 1/ l, (8) k where S stands for the complementary of S in {1,, n}. For an S and η being given, we partition S in the following way: S 1 := {indices of the l largest entries in absolute value of η in S}, S := {indices of the next l largest entries in absolute value of η in S},. Using one of the above inequalities We write η S + η S1 1 α l η S q k 1/q 1/ η S k 1/q 1/ η S + η S1. = 1 α l A(η S + η S1 ) = 1 α l [ ] A(η Sr ), A(η Sh ). h,r A( r η Sr ), A( h η Sh ) We renormalize η Sr, η Sh so that they have an l -norm equal to one by setting It follows that η Sr := η Sr / η Sr, η Sh := η Sh / η Sh for r, h. 1 A(η Sr ), A(η Sh ) = A( η Sr ), A( η Sh ) = 1 [ A( ηsr + η Sh ) η Sr η Sh 4 A( η S r η Sh ) ] 1 [ β 4 l η Sr + η Sh αl η Sr η Sh ] 1 = [β l αl]. 6

7 Thus, we have obtained A(η Sr ), A(η Sh ) β l α l η Sr η Sh. Consequently, we have A(η S + η S1 ) 1 [β l α l ] r,h η Sr η Sh = β l α l ( ) η Sh h and ( ) η S + η S1 1 A(η αl S + η S1 ) cond l(a) 1 η Sh. h Now consider an integer h. For i S h and j S h 1, the inequality η i η j yields, by raising to the power q and averaging over j, the inequality η i q 1 l η S h 1 q q, which in turns implies, by raising to the power /q and summing over i, the inequality η Sh l 1 /q η Sh 1 q. We derive h η Sh l 1/ 1/q h 1 [ 1/q η Sh q l 1/ 1/q η Sh q] q = l 1/ 1/q η S q. h 1 We then deduce η S + η S1 condl (A) 1 l 1/ 1/q η S q. In conclusion, we obtain the desired claim Eq. (8): η S q k 1/q 1/ η S k 1/q 1/ η S + η S1 condl (A) 1 ( l k) 1/ 1/q η S q. Step : l q -estimate. Let S be an index set corresponding to the k largest entries of x in absolute value. Note that x S minimizes x z q over all k-sparse vectors z, i.e., z 0 k. With x := q Ax, observe that η := x x belongs to the kernel of A. We have This implies x q q x q q = x S q q + x S q q x S q q η S q q + η S q q x S q q. η S q q η S q q + x S q q = η S q q + σ k(x) q q ρq η S q q + σ k(x) q q, hence η S q q 1 ρ qσ k(x) q q. 7

8 We finally derive η q q η S q q + η S q q (1 + ρq ) η S q q (1 + ρq ) 1 ρ q σ k (x) q q. Step 3: l -estimate. With η and S, S 1,... defined as before, we recall that η S S1 = η S + η S1 ρk 1/ 1/q η S q, ( ) 1/ η S S1 = η Sh η Sh k 1/ 1/q η S q. h We have now established the desired result: η = η S S1 + η S S1 (1 + ρ )k 1 /q η S q /q (1 + ρ ) (1 ρ q ) /q k1 /q σ k (x) q. This completes the proof. In fact, the result in Theorem 1.5 can be slightly improved. That is, letting c l = cond l (A), ( η S q ρ η S q with ρ cl 1 )( 1 + c l + 3 ) ( ) 1/q 1/ k := ( c l + 1 ). (9) l As in Eq. (8), we had an upper for η S + η S1. We now uncover a lower bound for η S + η S1. Indeed, since η = η S + η S1 + ker(a), η S1 1 A(η βl S1 ) = 1 A( η βl Sh ), A( η Sh ) h =1 h =1 = 1 [ A(ηS ), A(η βl S ) + A(η S ), A( η Sh ) + A( η Sh ), A( η Sh ) ] h h h 1 [ α βl l η S A( η 0 ), A(η Sh ) + A(η S + η S1 ) ]. h As before, we obtain A( η S ), A(η Sh ) β l α l η S η Sh. Thus, ( ) η S + η S α l η βl S β l α l η βl Sh η S + 1 A(η βl S + η S1 ). Using the estimates in the proof of Theorem 1.5, we may now write ( ( ) ( α l ) η S + 1 α l 1 η Sh η S + 1 β l β l h (1 + 1 ) η S (1 c + 1 ) Σ η S + 1 l c l h h α l (1 1 c l )( c l 1 β l ) Σ, ) A(η S + η S1 ) 8

9 where Σ = h η S h. Since the quadratic polynomial (c l + 1)x + (1 c l )Σ x 1 (c l 1) Σ takes a nonpositive value at x = η S, we derive It now remains to write η S (c l 1)Σ + (c l 1) Σ + (c l + 1)(c l 1) Σ (c l + 1) = (c l 1)Σ(1 + c l + 3) =: ρσ. (c l + 1) η S q k 1/q 1/ η S k 1/q 1/ ρσ k 1/q 1/ ρl 1/ 1/q η S q = ρ It follows that Theorem. Let c l = cond l (A). Suppose that ( cl 1 )( 1 + c l + 3 ) ρ := ( c l + 1 ) ( ) 1/q 1/ k < 1. l ( ) 1/q 1/ k η l S q. The solution of Eq. (3) is the sparsest solution. The estimates in Eq. (4) hold. Let us make the following remark. On the role of the null space property, the case q = 1 is a natural one. Given a vector x supported on S with #(S) k, we want to have x = argmin{ z 1, Az = Ax}, i.e. x 1 x η 1 for all η ker(a). Using a characterization of best l 1 -approximation, i.e. 0 being a best approximation to x from ker(a), the desired property is equivalent to η ker(a), sign(x i )η i η i. i S i S Since we require this to hold for any x supported on S, it is necessary and sufficient to ask η ker(a), η S 1 η S 1. As seen in the above proof, the slightly stronger condition η S 1 ρ η S 1 with ρ < 1 even yields instance optimality, not just exact recovery of sparse vectors. In fact, a very simple argument reveals that the null space property is also somewhat necessary in the l q -case. Indeed, if η belongs to ker(a), and if an index set S has cardinality at most k, then A(η S ) = A( η S ) and since η S should be recovered by l q -minimization, one should have η S q η S q. 9

10 We next follow the arguments in [Candes and Tao 05] to discuss the value cond k (A). Consider a matrix A of size m n whose entries are i.i.d. Gaussian with mean zero and variance 1/m. Let T be a subset of {1,,, n}. Following upon the work of Geman and Wachter as discussed in [Candes and Tao 05], we have (cf. [1]) and λ min (A T A T) (1 λ), a.s. λ max (A T A T) (1 + λ), a.s. (cf. [17]) as m and #(T) with #(T)/m λ < 1. Thus, c k = cond k (A) 1 + λ 1 λ, a.s. By Corollary,1, it follows that λ < 1/. That is, almost surely, when k = #(T) < m/4, the solution of Eq. () is the sparsest solution. Our numerical experiments in the last section verify this case. 3 Recovery From Imperfect Data We next consider the situation that the measurements y are imperfect. That is, y = Ax 0 +e with unknown perturbation vector e which is bounded by a known amount e η. In this case we consider the minimization for 0 < q < 1. min{ x q q, x R n, Ax y θ}. (10) A solution of the above minimization is denoted by q,θ y. When q = 1, we refer to [Donoho, Elad, and Temlyakov 06] ([14]) for results based on the mutual coherence of the columns of A and related literatures. As in the previous section, we have Theorem 3.1 Fix 0 < q < 1 and y. There exists a solution q,θ y solving Eq. (10) for every positive θ. Similar to the proof of Theorem 1.5 we can show Theorem 3. Fix q (0, 1]. Suppose that c l := cond l (A) < 1 + ( ) /q 1 l, k for some integer l > k. Then every k-sparse vector x is approximated by q,theta y, the minimization solution of Eq. (3) satisfying x q,θ y C(q, c l) k 1/q 1/σ k(x) q + C(k)θ, (11) where σ k (x) q := min z 0 k x z q, C(q, c l ) is a positive constant and C(k) is a positive constant dependent on α k. 10

11 Proof. Let q Ax be the solution of Eq. (3. Then Note that we have x q,θ y x q Ax + q,θ y q Ax q,θ y q Ax 1 α k A( q,θ y q Ax) = 1 α k A( q,θ y y 1 α k Combining with Theorem 1.5 we immediately obtain the desired result. 4 Numerical Computation In this section we propose an algorithm to compute the minimization problem Eq. (3). Since it is a nonconvex minimization, we have to approximate it. Fix a positive number ɛ > 0. Let n l q,ɛ (x) = x i ( x i + ɛ) q 1 be a functional of x R n. Instead of the minimization of Eq. (3), we consider i=1 min{l q,ɛ (x), x R n, Ax = y}. (1) for a fixed ɛ > 0 and q > 0. Similar to the proof of Theorem.1, we can see that the solution of the above minimization exists. We next show that the solution x ɛ of Eq.(1) will converge to the solution Eq. (3). For convenience, let x be a solution of Eq. (3). Theorem 4.1 Fix 0 < q 1. Let x ɛ be the solution of Eq. (1). Then x ɛ converges to one of the minimizers of Eq. (3) as ɛ 0 +. Proof. Let ɛ = 1/j and x j be the solution of the minimization problem Eq. (1) for ɛ = 1/j. Writing x j = (x j,1,, x j,n ) T, it is easy to see that x j,i are bounded for each j. Thus, there is a subsequence from {x j, j = 1,,, } which is convergent. For convenience, let us say x j converges to x. We claim that x is a solution in Eq. (3). Recall that x is a solution of Eq. (3). Clearly, l q,1/j (x j ) l q,1/j ( x) x q q x j q q. (13) It is easy to see that f ɛ (t) = t ( t + ɛ) q 1 are continuous functions for any ɛ > 0 and f ɛ (t) is decreasing for ɛ for ɛ > 0. By Dini s theorem, f ɛ (t) is uniformly convergent to f 0 (t) over a bounded interval [ B, B]. Note that t ( t + ɛ) q 1 t q 0 11

12 as ɛ 0 + uniformly for all t [ B, B]. Thus, we have x j q q l q,1/j(x j ) 0 as j. By using Eq. (13), it follows that x q = x q. This completes the proof. We now derive a simple algorithm to compute the minimization in Eq. (1). Let x 0 be any solution solving Ax 0 = y. Fix q > 0 and ɛ > 0. We solve the following minimization problem to find x j : min x 1,,x n { n i=1 x i ( x (j 1) i + ɛ) q 1, Ax = y} (14) for j = 1,,,. If x j x j 1 1 is very small enough, we stop the iterations. This method is very easy to implement, e.g., by using the l 1 method to a weighted matrix Ãj of A. More precisely, letting ˆx = (ˆx 1,, ˆx n ) T with ˆx i = x i ( x (j 1) i + ɛ) q 1 and Âj = AD j with D being a diagonal matrix with entries ( x (j 1) i + ɛ) 1 q. That is, we use an l 1 minimization algorithm to solve min{ ˆx 1, Â jˆx = y} to get ˆx, and hence, x (j). It is interesting to point out that when q = 0, this algorithm becomes the reweighted l 1 method discussed in [6]. However, the algorithm proposed above may not converge for some q. A steepest descent method with some variations for the minimization problem Eq. (1) was extensively experimented in [8]. In the following section we use our simple algorithm to do some modest experiments. Numerical experiments show that l q minimization can work much better. 5 Numerical Experiments We now compare our method with several existing methods: l 1 method, reweighted l 1 mehtod, the OGA (cf. [0]), ROMP(cf. [19]) and our simple iterative l q method. Let us describe our comparisons in detail as follows. Example 1. Fix an integer 1 k 60. We choose a matrix A of size with i.i.d. Gaussian random entrices with zero mean and variance 1. For each A, we use a vector p of size 51 1 which contains a random permutation of integers 1,,, 51. For an index set T = q(1 : k), we let x(t) = randn(1 : k)) (in MATLAB command) be the true solution with x 0 k, we let y = Ax be the given vector. Choose x 0 = A(1 : 18, 1 : 18) 1 y and pad x 0 by zeros to form an initial guess vector of size Then we use our new method as well as the existing l 1 method, the reweighted l 1 method, OGA method described in [0] (The authors thank Alex Petukhov for providing his MATLAB program for comparison) and the ROMP method to find x with given A, y, x 0 and accuracy For our method we vary the choices of 0 < ɛ < 1 and q < 1. If x x < 10 3, we assume that the method is successful, otherwise the method fails. We repeat the above experiment 1

13 Comparison of l 1, rwl 1, OGA, l q, ROMP methods l 1 0. rw l 1 OGA l q ROMP Figure 1: Comparison of l 1 and l q methods for sparest solutions 100 times for each k from 1 to 60 and record the number of success that a method can find the solution during the 100 experiment. We divide these numbers by 100 and plot the pecentages of successfully solving the solutions by the 5 methods in the following figure From Figure 1, we can see that our new method does provide a better way to solve the Eq. (1) problem than all the other methods. Examle. In addition, we performed another set of comparion. Our numerical experiment is exactly the same as above except for x(t) = sign(randn)(1 : k) to the exact sparsest vector. The percentages of successfully solving the sparsest solutions by the methods mentioned in Example 1 are shown in Fig.. The figure above again shows that our method is the best among the all methods discussed in this paper. References [1] Candès, E. J. Compressive sampling. International Congress of Mathematicians. Vol. III, , Eur. Math. Soc., Z urich, 006. [] Candès, E. J., J. K. Romberg, Quantitative robust uncertainty principles and optimally sparse decompositions. Found. Comput. Math. 6 (006),, [3] Candès, E. J., J. K. Romberg, J. K. and T. Tao, Stable signal recovery from incomplete and inaccurate measurements. Comm. Pure Appl. Math. 59 (006),

14 Comparison of l 1, rwl 1, OGA, l q, ROMP methods 1 l 1 rw l 1 OGA l q ROMP Figure : Comparison of l 1, l q, and OGA methods for sparest solutions [4] Candès, E. J. and T. Tao, Decoding by linear programming, IEEE Trans. Inform. Theory 51 (005), no. 1, [5] Candès, E. J. and T. Tao, Near-optimal signal recovery from random projections: universal encoding strategies, IEEE Trans. Inform. Theory 5 (006), no. 1, [6] Candès, E. J., M. Watkin, and S. Boyd, Enhancing Sparsity by Reweighted l 1 Minimization, manuscritp, 007. [7] R. Chartrand, Nonconvex compressed sensing and error correction, in International Conference on Acoustics, Speech, and Signal Processing, IEEE, 007. [8] R. Chartrans, Exact reconstruction of sparse signals via nonconvex minimization, manuscript, 007. [9] Donoho, D. L., Compressed sensing, IEEE Trans. Inform. Theory 5 (006), [10] Donoho, D. L., Sparse components of images and optimal atomic decompositions. Constr. Approx. 17 (001), [11] D. L. Donoho, Unconditional bases are optimal bases for data compression and for statistical estimation, Appl. Comput. Harmonic Anal., 1(1993),

15 [1] D. L. Donoho, Unconditional bases and bit-level compression, Appl. Comput. Harmonic Anal., 3(1996), pp [13] Donoho, D. L. and M. Elad, Optimally sparse representation in general (nonorthogonal) dictionaries via l 1 minimization, Proc. Natl. Acad. Sci. USA 100 (003), no. 5, [14] Donoho, D. L., M. Elad, and V. N. Temlyakov, Stable recovery of sparse overcomplete representations in the presence of noise. IEEE Trans. Inform. Theory, 5 (006), [15] Donoho, D. L. and J. Tanner, Sparse nonnegative solution of underdetermined linear equations by linear programming. Proc. Natl. Acad. Sci. USA 10 (005), no. 7, [16] Elad, M. and A. M. Bruckstein, IEEE Trans. Inf. Theory 48(00), [17] Geman, S., A limit theorem for the norm of random matrices, Ann. Prob., 8(1980), [18] Natarajan, B. K., Sparse approximate solutions to linear systems, SIAM J. Comput., vol. 4, pp. 734, [19] Needell, D. and R. Vershynin, Uniform uncertainty principal and signal recovery via regularized orthogonal matching pursuit, manuscript, 007. [0] Petukhov, A., Fast implementation of orthogonal greedy algorithm for tight wavelet frames, Signal Processing, 86(006), [1] Wachter, K. W., The strong limits of random matrix spectra for sample matrices of independent elements, Ann. Prob., 6(1978),

Sparsest Solutions of Underdetermined Linear Systems via l q -minimization for 0 < q 1

Sparsest Solutions of Underdetermined Linear Systems via l q -minimization for 0 < q 1 Sparsest Solutions of Underdetermined Linear Systems via l q -minimization for 0 < q 1 Simon Foucart Department of Mathematics Vanderbilt University Nashville, TN 3740 Ming-Jun Lai Department of Mathematics

More information

This article appeared in a journal published by Elsevier. The attached copy is furnished to the author for internal non-commercial research and

This article appeared in a journal published by Elsevier. The attached copy is furnished to the author for internal non-commercial research and This article appeared in a journal published by Elsevier The attached copy is furnished to the author for internal non-commercial research and education use, including for instruction at the authors institution

More information

On Sparse Solutions of Underdetermined Linear Systems

On Sparse Solutions of Underdetermined Linear Systems On Sparse Solutions of Underdetermined Linear Systems Ming-Jun Lai Department of Mathematics the University of Georgia Athens, GA 30602 January 17, 2009 Abstract We first explain the research problem of

More information

An Unconstrained l q Minimization with 0 < q 1 for Sparse Solution of Under-determined Linear Systems

An Unconstrained l q Minimization with 0 < q 1 for Sparse Solution of Under-determined Linear Systems An Unconstrained l q Minimization with 0 < q 1 for Sparse Solution of Under-determined Linear Systems Ming-Jun Lai and Jingyue Wang Department of Mathematics The University of Georgia Athens, GA 30602.

More information

New Coherence and RIP Analysis for Weak. Orthogonal Matching Pursuit

New Coherence and RIP Analysis for Weak. Orthogonal Matching Pursuit New Coherence and RIP Analysis for Wea 1 Orthogonal Matching Pursuit Mingrui Yang, Member, IEEE, and Fran de Hoog arxiv:1405.3354v1 [cs.it] 14 May 2014 Abstract In this paper we define a new coherence

More information

Uniqueness Conditions for A Class of l 0 -Minimization Problems

Uniqueness Conditions for A Class of l 0 -Minimization Problems Uniqueness Conditions for A Class of l 0 -Minimization Problems Chunlei Xu and Yun-Bin Zhao October, 03, Revised January 04 Abstract. We consider a class of l 0 -minimization problems, which is to search

More information

of Orthogonal Matching Pursuit

of Orthogonal Matching Pursuit A Sharp Restricted Isometry Constant Bound of Orthogonal Matching Pursuit Qun Mo arxiv:50.0708v [cs.it] 8 Jan 205 Abstract We shall show that if the restricted isometry constant (RIC) δ s+ (A) of the measurement

More information

RSP-Based Analysis for Sparsest and Least l 1 -Norm Solutions to Underdetermined Linear Systems

RSP-Based Analysis for Sparsest and Least l 1 -Norm Solutions to Underdetermined Linear Systems 1 RSP-Based Analysis for Sparsest and Least l 1 -Norm Solutions to Underdetermined Linear Systems Yun-Bin Zhao IEEE member Abstract Recently, the worse-case analysis, probabilistic analysis and empirical

More information

Lecture: Introduction to Compressed Sensing Sparse Recovery Guarantees

Lecture: Introduction to Compressed Sensing Sparse Recovery Guarantees Lecture: Introduction to Compressed Sensing Sparse Recovery Guarantees http://bicmr.pku.edu.cn/~wenzw/bigdata2018.html Acknowledgement: this slides is based on Prof. Emmanuel Candes and Prof. Wotao Yin

More information

A new method on deterministic construction of the measurement matrix in compressed sensing

A new method on deterministic construction of the measurement matrix in compressed sensing A new method on deterministic construction of the measurement matrix in compressed sensing Qun Mo 1 arxiv:1503.01250v1 [cs.it] 4 Mar 2015 Abstract Construction on the measurement matrix A is a central

More information

Thresholds for the Recovery of Sparse Solutions via L1 Minimization

Thresholds for the Recovery of Sparse Solutions via L1 Minimization Thresholds for the Recovery of Sparse Solutions via L Minimization David L. Donoho Department of Statistics Stanford University 39 Serra Mall, Sequoia Hall Stanford, CA 9435-465 Email: donoho@stanford.edu

More information

Stable Signal Recovery from Incomplete and Inaccurate Measurements

Stable Signal Recovery from Incomplete and Inaccurate Measurements Stable Signal Recovery from Incomplete and Inaccurate Measurements EMMANUEL J. CANDÈS California Institute of Technology JUSTIN K. ROMBERG California Institute of Technology AND TERENCE TAO University

More information

GREEDY SIGNAL RECOVERY REVIEW

GREEDY SIGNAL RECOVERY REVIEW GREEDY SIGNAL RECOVERY REVIEW DEANNA NEEDELL, JOEL A. TROPP, ROMAN VERSHYNIN Abstract. The two major approaches to sparse recovery are L 1-minimization and greedy methods. Recently, Needell and Vershynin

More information

Sparse Optimization Lecture: Sparse Recovery Guarantees

Sparse Optimization Lecture: Sparse Recovery Guarantees Those who complete this lecture will know Sparse Optimization Lecture: Sparse Recovery Guarantees Sparse Optimization Lecture: Sparse Recovery Guarantees Instructor: Wotao Yin Department of Mathematics,

More information

A New Estimate of Restricted Isometry Constants for Sparse Solutions

A New Estimate of Restricted Isometry Constants for Sparse Solutions A New Estimate of Restricted Isometry Constants for Sparse Solutions Ming-Jun Lai and Louis Y. Liu January 12, 211 Abstract We show that as long as the restricted isometry constant δ 2k < 1/2, there exist

More information

Rui ZHANG Song LI. Department of Mathematics, Zhejiang University, Hangzhou , P. R. China

Rui ZHANG Song LI. Department of Mathematics, Zhejiang University, Hangzhou , P. R. China Acta Mathematica Sinica, English Series May, 015, Vol. 31, No. 5, pp. 755 766 Published online: April 15, 015 DOI: 10.1007/s10114-015-434-4 Http://www.ActaMath.com Acta Mathematica Sinica, English Series

More information

Near Ideal Behavior of a Modified Elastic Net Algorithm in Compressed Sensing

Near Ideal Behavior of a Modified Elastic Net Algorithm in Compressed Sensing Near Ideal Behavior of a Modified Elastic Net Algorithm in Compressed Sensing M. Vidyasagar Cecil & Ida Green Chair The University of Texas at Dallas M.Vidyasagar@utdallas.edu www.utdallas.edu/ m.vidyasagar

More information

Signal Recovery From Incomplete and Inaccurate Measurements via Regularized Orthogonal Matching Pursuit

Signal Recovery From Incomplete and Inaccurate Measurements via Regularized Orthogonal Matching Pursuit Signal Recovery From Incomplete and Inaccurate Measurements via Regularized Orthogonal Matching Pursuit Deanna Needell and Roman Vershynin Abstract We demonstrate a simple greedy algorithm that can reliably

More information

Generalized Orthogonal Matching Pursuit- A Review and Some

Generalized Orthogonal Matching Pursuit- A Review and Some Generalized Orthogonal Matching Pursuit- A Review and Some New Results Department of Electronics and Electrical Communication Engineering Indian Institute of Technology, Kharagpur, INDIA Table of Contents

More information

INDUSTRIAL MATHEMATICS INSTITUTE. B.S. Kashin and V.N. Temlyakov. IMI Preprint Series. Department of Mathematics University of South Carolina

INDUSTRIAL MATHEMATICS INSTITUTE. B.S. Kashin and V.N. Temlyakov. IMI Preprint Series. Department of Mathematics University of South Carolina INDUSTRIAL MATHEMATICS INSTITUTE 2007:08 A remark on compressed sensing B.S. Kashin and V.N. Temlyakov IMI Preprint Series Department of Mathematics University of South Carolina A remark on compressed

More information

Stability and Robustness of Weak Orthogonal Matching Pursuits

Stability and Robustness of Weak Orthogonal Matching Pursuits Stability and Robustness of Weak Orthogonal Matching Pursuits Simon Foucart, Drexel University Abstract A recent result establishing, under restricted isometry conditions, the success of sparse recovery

More information

Exact Low-rank Matrix Recovery via Nonconvex M p -Minimization

Exact Low-rank Matrix Recovery via Nonconvex M p -Minimization Exact Low-rank Matrix Recovery via Nonconvex M p -Minimization Lingchen Kong and Naihua Xiu Department of Applied Mathematics, Beijing Jiaotong University, Beijing, 100044, People s Republic of China E-mail:

More information

Error Correction via Linear Programming

Error Correction via Linear Programming Error Correction via Linear Programming Emmanuel Candes and Terence Tao Applied and Computational Mathematics, Caltech, Pasadena, CA 91125 Department of Mathematics, University of California, Los Angeles,

More information

Sparse Solutions of Linear Systems of Equations and Sparse Modeling of Signals and Images!

Sparse Solutions of Linear Systems of Equations and Sparse Modeling of Signals and Images! Sparse Solutions of Linear Systems of Equations and Sparse Modeling of Signals and Images! Alfredo Nava-Tudela John J. Benedetto, advisor 1 Happy birthday Lucía! 2 Outline - Problem: Find sparse solutions

More information

Inverse problems and sparse models (1/2) Rémi Gribonval INRIA Rennes - Bretagne Atlantique, France

Inverse problems and sparse models (1/2) Rémi Gribonval INRIA Rennes - Bretagne Atlantique, France Inverse problems and sparse models (1/2) Rémi Gribonval INRIA Rennes - Bretagne Atlantique, France remi.gribonval@inria.fr Structure of the tutorial Session 1: Introduction to inverse problems & sparse

More information

Compressed Sensing and Sparse Recovery

Compressed Sensing and Sparse Recovery ELE 538B: Sparsity, Structure and Inference Compressed Sensing and Sparse Recovery Yuxin Chen Princeton University, Spring 217 Outline Restricted isometry property (RIP) A RIPless theory Compressed sensing

More information

An Introduction to Sparse Approximation

An Introduction to Sparse Approximation An Introduction to Sparse Approximation Anna C. Gilbert Department of Mathematics University of Michigan Basic image/signal/data compression: transform coding Approximate signals sparsely Compress images,

More information

Greedy Signal Recovery and Uniform Uncertainty Principles

Greedy Signal Recovery and Uniform Uncertainty Principles Greedy Signal Recovery and Uniform Uncertainty Principles SPIE - IE 2008 Deanna Needell Joint work with Roman Vershynin UC Davis, January 2008 Greedy Signal Recovery and Uniform Uncertainty Principles

More information

Orthogonal Matching Pursuit for Sparse Signal Recovery With Noise

Orthogonal Matching Pursuit for Sparse Signal Recovery With Noise Orthogonal Matching Pursuit for Sparse Signal Recovery With Noise The MIT Faculty has made this article openly available. Please share how this access benefits you. Your story matters. Citation As Published

More information

Uniform Uncertainty Principle and signal recovery via Regularized Orthogonal Matching Pursuit

Uniform Uncertainty Principle and signal recovery via Regularized Orthogonal Matching Pursuit Uniform Uncertainty Principle and signal recovery via Regularized Orthogonal Matching Pursuit arxiv:0707.4203v2 [math.na] 14 Aug 2007 Deanna Needell Department of Mathematics University of California,

More information

Introduction to Compressed Sensing

Introduction to Compressed Sensing Introduction to Compressed Sensing Alejandro Parada, Gonzalo Arce University of Delaware August 25, 2016 Motivation: Classical Sampling 1 Motivation: Classical Sampling Issues Some applications Radar Spectral

More information

Uniform Uncertainty Principle and Signal Recovery via Regularized Orthogonal Matching Pursuit

Uniform Uncertainty Principle and Signal Recovery via Regularized Orthogonal Matching Pursuit Claremont Colleges Scholarship @ Claremont CMC Faculty Publications and Research CMC Faculty Scholarship 6-5-2008 Uniform Uncertainty Principle and Signal Recovery via Regularized Orthogonal Matching Pursuit

More information

Recent Developments in Compressed Sensing

Recent Developments in Compressed Sensing Recent Developments in Compressed Sensing M. Vidyasagar Distinguished Professor, IIT Hyderabad m.vidyasagar@iith.ac.in, www.iith.ac.in/ m vidyasagar/ ISL Seminar, Stanford University, 19 April 2018 Outline

More information

IEEE TRANSACTIONS ON INFORMATION THEORY, VOL. 57, NO. 11, NOVEMBER On the Performance of Sparse Recovery

IEEE TRANSACTIONS ON INFORMATION THEORY, VOL. 57, NO. 11, NOVEMBER On the Performance of Sparse Recovery IEEE TRANSACTIONS ON INFORMATION THEORY, VOL. 57, NO. 11, NOVEMBER 2011 7255 On the Performance of Sparse Recovery Via `p-minimization (0 p 1) Meng Wang, Student Member, IEEE, Weiyu Xu, and Ao Tang, Senior

More information

IEEE SIGNAL PROCESSING LETTERS, VOL. 22, NO. 9, SEPTEMBER

IEEE SIGNAL PROCESSING LETTERS, VOL. 22, NO. 9, SEPTEMBER IEEE SIGNAL PROCESSING LETTERS, VOL. 22, NO. 9, SEPTEMBER 2015 1239 Preconditioning for Underdetermined Linear Systems with Sparse Solutions Evaggelia Tsiligianni, StudentMember,IEEE, Lisimachos P. Kondi,

More information

Noisy Signal Recovery via Iterative Reweighted L1-Minimization

Noisy Signal Recovery via Iterative Reweighted L1-Minimization Noisy Signal Recovery via Iterative Reweighted L1-Minimization Deanna Needell UC Davis / Stanford University Asilomar SSC, November 2009 Problem Background Setup 1 Suppose x is an unknown signal in R d.

More information

Constructing Explicit RIP Matrices and the Square-Root Bottleneck

Constructing Explicit RIP Matrices and the Square-Root Bottleneck Constructing Explicit RIP Matrices and the Square-Root Bottleneck Ryan Cinoman July 18, 2018 Ryan Cinoman Constructing Explicit RIP Matrices July 18, 2018 1 / 36 Outline 1 Introduction 2 Restricted Isometry

More information

Sparse Solutions of Linear Systems of Equations and Sparse Modeling of Signals and Images: Final Presentation

Sparse Solutions of Linear Systems of Equations and Sparse Modeling of Signals and Images: Final Presentation Sparse Solutions of Linear Systems of Equations and Sparse Modeling of Signals and Images: Final Presentation Alfredo Nava-Tudela John J. Benedetto, advisor 5/10/11 AMSC 663/664 1 Problem Let A be an n

More information

CoSaMP: Greedy Signal Recovery and Uniform Uncertainty Principles

CoSaMP: Greedy Signal Recovery and Uniform Uncertainty Principles CoSaMP: Greedy Signal Recovery and Uniform Uncertainty Principles SIAM Student Research Conference Deanna Needell Joint work with Roman Vershynin and Joel Tropp UC Davis, May 2008 CoSaMP: Greedy Signal

More information

AN INTRODUCTION TO COMPRESSIVE SENSING

AN INTRODUCTION TO COMPRESSIVE SENSING AN INTRODUCTION TO COMPRESSIVE SENSING Rodrigo B. Platte School of Mathematical and Statistical Sciences APM/EEE598 Reverse Engineering of Complex Dynamical Networks OUTLINE 1 INTRODUCTION 2 INCOHERENCE

More information

Sparse Recovery with Pre-Gaussian Random Matrices

Sparse Recovery with Pre-Gaussian Random Matrices Sparse Recovery with Pre-Gaussian Random Matrices Simon Foucart Laboratoire Jacques-Louis Lions Université Pierre et Marie Curie Paris, 75013, France Ming-Jun Lai Department of Mathematics University of

More information

Sensing systems limited by constraints: physical size, time, cost, energy

Sensing systems limited by constraints: physical size, time, cost, energy Rebecca Willett Sensing systems limited by constraints: physical size, time, cost, energy Reduce the number of measurements needed for reconstruction Higher accuracy data subject to constraints Original

More information

Recovering overcomplete sparse representations from structured sensing

Recovering overcomplete sparse representations from structured sensing Recovering overcomplete sparse representations from structured sensing Deanna Needell Claremont McKenna College Feb. 2015 Support: Alfred P. Sloan Foundation and NSF CAREER #1348721. Joint work with Felix

More information

Necessary and sufficient conditions of solution uniqueness in l 1 minimization

Necessary and sufficient conditions of solution uniqueness in l 1 minimization 1 Necessary and sufficient conditions of solution uniqueness in l 1 minimization Hui Zhang, Wotao Yin, and Lizhi Cheng arxiv:1209.0652v2 [cs.it] 18 Sep 2012 Abstract This paper shows that the solutions

More information

Stability and robustness of l 1 -minimizations with Weibull matrices and redundant dictionaries

Stability and robustness of l 1 -minimizations with Weibull matrices and redundant dictionaries Stability and robustness of l 1 -minimizations with Weibull matrices and redundant dictionaries Simon Foucart, Drexel University Abstract We investigate the recovery of almost s-sparse vectors x C N from

More information

Pre-weighted Matching Pursuit Algorithms for Sparse Recovery

Pre-weighted Matching Pursuit Algorithms for Sparse Recovery Journal of Information & Computational Science 11:9 (214) 2933 2939 June 1, 214 Available at http://www.joics.com Pre-weighted Matching Pursuit Algorithms for Sparse Recovery Jingfei He, Guiling Sun, Jie

More information

Conditions for Robust Principal Component Analysis

Conditions for Robust Principal Component Analysis Rose-Hulman Undergraduate Mathematics Journal Volume 12 Issue 2 Article 9 Conditions for Robust Principal Component Analysis Michael Hornstein Stanford University, mdhornstein@gmail.com Follow this and

More information

Course Notes for EE227C (Spring 2018): Convex Optimization and Approximation

Course Notes for EE227C (Spring 2018): Convex Optimization and Approximation Course Notes for EE227C (Spring 2018): Convex Optimization and Approximation Instructor: Moritz Hardt Email: hardt+ee227c@berkeley.edu Graduate Instructor: Max Simchowitz Email: msimchow+ee227c@berkeley.edu

More information

Multipath Matching Pursuit

Multipath Matching Pursuit Multipath Matching Pursuit Submitted to IEEE trans. on Information theory Authors: S. Kwon, J. Wang, and B. Shim Presenter: Hwanchol Jang Multipath is investigated rather than a single path for a greedy

More information

Reconstruction from Anisotropic Random Measurements

Reconstruction from Anisotropic Random Measurements Reconstruction from Anisotropic Random Measurements Mark Rudelson and Shuheng Zhou The University of Michigan, Ann Arbor Coding, Complexity, and Sparsity Workshop, 013 Ann Arbor, Michigan August 7, 013

More information

A Note on the Complexity of L p Minimization

A Note on the Complexity of L p Minimization Mathematical Programming manuscript No. (will be inserted by the editor) A Note on the Complexity of L p Minimization Dongdong Ge Xiaoye Jiang Yinyu Ye Abstract We discuss the L p (0 p < 1) minimization

More information

Introduction to Sparsity. Xudong Cao, Jake Dreamtree & Jerry 04/05/2012

Introduction to Sparsity. Xudong Cao, Jake Dreamtree & Jerry 04/05/2012 Introduction to Sparsity Xudong Cao, Jake Dreamtree & Jerry 04/05/2012 Outline Understanding Sparsity Total variation Compressed sensing(definition) Exact recovery with sparse prior(l 0 ) l 1 relaxation

More information

Lecture 22: More On Compressed Sensing

Lecture 22: More On Compressed Sensing Lecture 22: More On Compressed Sensing Scribed by Eric Lee, Chengrun Yang, and Sebastian Ament Nov. 2, 207 Recap and Introduction Basis pursuit was the method of recovering the sparsest solution to an

More information

Shifting Inequality and Recovery of Sparse Signals

Shifting Inequality and Recovery of Sparse Signals Shifting Inequality and Recovery of Sparse Signals T. Tony Cai Lie Wang and Guangwu Xu Astract In this paper we present a concise and coherent analysis of the constrained l 1 minimization method for stale

More information

Lecture Notes 9: Constrained Optimization

Lecture Notes 9: Constrained Optimization Optimization-based data analysis Fall 017 Lecture Notes 9: Constrained Optimization 1 Compressed sensing 1.1 Underdetermined linear inverse problems Linear inverse problems model measurements of the form

More information

COMPRESSED SENSING IN PYTHON

COMPRESSED SENSING IN PYTHON COMPRESSED SENSING IN PYTHON Sercan Yıldız syildiz@samsi.info February 27, 2017 OUTLINE A BRIEF INTRODUCTION TO COMPRESSED SENSING A BRIEF INTRODUCTION TO CVXOPT EXAMPLES A Brief Introduction to Compressed

More information

Abstract This paper is about the efficient solution of large-scale compressed sensing problems.

Abstract This paper is about the efficient solution of large-scale compressed sensing problems. Noname manuscript No. (will be inserted by the editor) Optimization for Compressed Sensing: New Insights and Alternatives Robert Vanderbei and Han Liu and Lie Wang Received: date / Accepted: date Abstract

More information

A Generalized Uncertainty Principle and Sparse Representation in Pairs of Bases

A Generalized Uncertainty Principle and Sparse Representation in Pairs of Bases 2558 IEEE TRANSACTIONS ON INFORMATION THEORY, VOL 48, NO 9, SEPTEMBER 2002 A Generalized Uncertainty Principle Sparse Representation in Pairs of Bases Michael Elad Alfred M Bruckstein Abstract An elementary

More information

Observability of a Linear System Under Sparsity Constraints

Observability of a Linear System Under Sparsity Constraints 2372 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL 58, NO 9, SEPTEMBER 2013 Observability of a Linear System Under Sparsity Constraints Wei Dai and Serdar Yüksel Abstract Consider an -dimensional linear

More information

Necessary and Sufficient Conditions of Solution Uniqueness in 1-Norm Minimization

Necessary and Sufficient Conditions of Solution Uniqueness in 1-Norm Minimization Noname manuscript No. (will be inserted by the editor) Necessary and Sufficient Conditions of Solution Uniqueness in 1-Norm Minimization Hui Zhang Wotao Yin Lizhi Cheng Received: / Accepted: Abstract This

More information

ORTHOGONAL matching pursuit (OMP) is the canonical

ORTHOGONAL matching pursuit (OMP) is the canonical IEEE TRANSACTIONS ON INFORMATION THEORY, VOL. 56, NO. 9, SEPTEMBER 2010 4395 Analysis of Orthogonal Matching Pursuit Using the Restricted Isometry Property Mark A. Davenport, Member, IEEE, and Michael

More information

Least squares regularized or constrained by L0: relationship between their global minimizers. Mila Nikolova

Least squares regularized or constrained by L0: relationship between their global minimizers. Mila Nikolova Least squares regularized or constrained by L0: relationship between their global minimizers Mila Nikolova CMLA, CNRS, ENS Cachan, Université Paris-Saclay, France nikolova@cmla.ens-cachan.fr SIAM Minisymposium

More information

Introduction How it works Theory behind Compressed Sensing. Compressed Sensing. Huichao Xue. CS3750 Fall 2011

Introduction How it works Theory behind Compressed Sensing. Compressed Sensing. Huichao Xue. CS3750 Fall 2011 Compressed Sensing Huichao Xue CS3750 Fall 2011 Table of Contents Introduction From News Reports Abstract Definition How it works A review of L 1 norm The Algorithm Backgrounds for underdetermined linear

More information

Compressed Sensing and Robust Recovery of Low Rank Matrices

Compressed Sensing and Robust Recovery of Low Rank Matrices Compressed Sensing and Robust Recovery of Low Rank Matrices M. Fazel, E. Candès, B. Recht, P. Parrilo Electrical Engineering, University of Washington Applied and Computational Mathematics Dept., Caltech

More information

Strengthened Sobolev inequalities for a random subspace of functions

Strengthened Sobolev inequalities for a random subspace of functions Strengthened Sobolev inequalities for a random subspace of functions Rachel Ward University of Texas at Austin April 2013 2 Discrete Sobolev inequalities Proposition (Sobolev inequality for discrete images)

More information

Tractable Upper Bounds on the Restricted Isometry Constant

Tractable Upper Bounds on the Restricted Isometry Constant Tractable Upper Bounds on the Restricted Isometry Constant Alex d Aspremont, Francis Bach, Laurent El Ghaoui Princeton University, École Normale Supérieure, U.C. Berkeley. Support from NSF, DHS and Google.

More information

Sparse Solutions of an Undetermined Linear System

Sparse Solutions of an Undetermined Linear System 1 Sparse Solutions of an Undetermined Linear System Maddullah Almerdasy New York University Tandon School of Engineering arxiv:1702.07096v1 [math.oc] 23 Feb 2017 Abstract This work proposes a research

More information

Solution-recovery in l 1 -norm for non-square linear systems: deterministic conditions and open questions

Solution-recovery in l 1 -norm for non-square linear systems: deterministic conditions and open questions Solution-recovery in l 1 -norm for non-square linear systems: deterministic conditions and open questions Yin Zhang Technical Report TR05-06 Department of Computational and Applied Mathematics Rice University,

More information

Sparse signals recovered by non-convex penalty in quasi-linear systems

Sparse signals recovered by non-convex penalty in quasi-linear systems Cui et al. Journal of Inequalities and Applications 018) 018:59 https://doi.org/10.1186/s13660-018-165-8 R E S E A R C H Open Access Sparse signals recovered by non-conve penalty in quasi-linear systems

More information

Compressibility of Infinite Sequences and its Interplay with Compressed Sensing Recovery

Compressibility of Infinite Sequences and its Interplay with Compressed Sensing Recovery Compressibility of Infinite Sequences and its Interplay with Compressed Sensing Recovery Jorge F. Silva and Eduardo Pavez Department of Electrical Engineering Information and Decision Systems Group Universidad

More information

Sparse Solutions of Systems of Equations and Sparse Modelling of Signals and Images

Sparse Solutions of Systems of Equations and Sparse Modelling of Signals and Images Sparse Solutions of Systems of Equations and Sparse Modelling of Signals and Images Alfredo Nava-Tudela ant@umd.edu John J. Benedetto Department of Mathematics jjb@umd.edu Abstract In this project we are

More information

Two Results on the Schatten p-quasi-norm Minimization for Low-Rank Matrix Recovery

Two Results on the Schatten p-quasi-norm Minimization for Low-Rank Matrix Recovery Two Results on the Schatten p-quasi-norm Minimization for Low-Rank Matrix Recovery Ming-Jun Lai, Song Li, Louis Y. Liu and Huimin Wang August 14, 2012 Abstract We shall provide a sufficient condition to

More information

Optimization for Compressed Sensing

Optimization for Compressed Sensing Optimization for Compressed Sensing Robert J. Vanderbei 2014 March 21 Dept. of Industrial & Systems Engineering University of Florida http://www.princeton.edu/ rvdb Lasso Regression The problem is to solve

More information

CS 229r: Algorithms for Big Data Fall Lecture 19 Nov 5

CS 229r: Algorithms for Big Data Fall Lecture 19 Nov 5 CS 229r: Algorithms for Big Data Fall 215 Prof. Jelani Nelson Lecture 19 Nov 5 Scribe: Abdul Wasay 1 Overview In the last lecture, we started discussing the problem of compressed sensing where we are given

More information

Inverse problems and sparse models (6/6) Rémi Gribonval INRIA Rennes - Bretagne Atlantique, France.

Inverse problems and sparse models (6/6) Rémi Gribonval INRIA Rennes - Bretagne Atlantique, France. Inverse problems and sparse models (6/6) Rémi Gribonval INRIA Rennes - Bretagne Atlantique, France remi.gribonval@inria.fr Overview of the course Introduction sparsity & data compression inverse problems

More information

Recovery of Sparse Signals Using Multiple Orthogonal Least Squares

Recovery of Sparse Signals Using Multiple Orthogonal Least Squares Recovery of Sparse Signals Using Multiple Orthogonal east Squares Jian Wang, Ping i Department of Statistics and Biostatistics arxiv:40.505v [stat.me] 9 Oct 04 Department of Computer Science Rutgers University

More information

Conditions for a Unique Non-negative Solution to an Underdetermined System

Conditions for a Unique Non-negative Solution to an Underdetermined System Conditions for a Unique Non-negative Solution to an Underdetermined System Meng Wang and Ao Tang School of Electrical and Computer Engineering Cornell University Ithaca, NY 14853 Abstract This paper investigates

More information

Analysis of Greedy Algorithms

Analysis of Greedy Algorithms Analysis of Greedy Algorithms Jiahui Shen Florida State University Oct.26th Outline Introduction Regularity condition Analysis on orthogonal matching pursuit Analysis on forward-backward greedy algorithm

More information

1-Bit Compressive Sensing

1-Bit Compressive Sensing 1-Bit Compressive Sensing Petros T. Boufounos, Richard G. Baraniuk Rice University, Electrical and Computer Engineering 61 Main St. MS 38, Houston, TX 775 Abstract Compressive sensing is a new signal acquisition

More information

SPARSE signal representations have gained popularity in recent

SPARSE signal representations have gained popularity in recent 6958 IEEE TRANSACTIONS ON INFORMATION THEORY, VOL. 57, NO. 10, OCTOBER 2011 Blind Compressed Sensing Sivan Gleichman and Yonina C. Eldar, Senior Member, IEEE Abstract The fundamental principle underlying

More information

Multiplicative and Additive Perturbation Effects on the Recovery of Sparse Signals on the Sphere using Compressed Sensing

Multiplicative and Additive Perturbation Effects on the Recovery of Sparse Signals on the Sphere using Compressed Sensing Multiplicative and Additive Perturbation Effects on the Recovery of Sparse Signals on the Sphere using Compressed Sensing ibeltal F. Alem, Daniel H. Chae, and Rodney A. Kennedy The Australian National

More information

Sigma Delta Quantization for Compressed Sensing

Sigma Delta Quantization for Compressed Sensing Sigma Delta Quantization for Compressed Sensing C. Sinan Güntürk, 1 Mark Lammers, 2 Alex Powell, 3 Rayan Saab, 4 Özgür Yılmaz 4 1 Courant Institute of Mathematical Sciences, New York University, NY, USA.

More information

Constrained optimization

Constrained optimization Constrained optimization DS-GA 1013 / MATH-GA 2824 Optimization-based Data Analysis http://www.cims.nyu.edu/~cfgranda/pages/obda_fall17/index.html Carlos Fernandez-Granda Compressed sensing Convex constrained

More information

Signal Recovery, Uncertainty Relations, and Minkowski Dimension

Signal Recovery, Uncertainty Relations, and Minkowski Dimension Signal Recovery, Uncertainty Relations, and Minkowski Dimension Helmut Bőlcskei ETH Zurich December 2013 Joint work with C. Aubel, P. Kuppinger, G. Pope, E. Riegler, D. Stotz, and C. Studer Aim of this

More information

Interpolation via weighted l 1 -minimization

Interpolation via weighted l 1 -minimization Interpolation via weighted l 1 -minimization Holger Rauhut RWTH Aachen University Lehrstuhl C für Mathematik (Analysis) Mathematical Analysis and Applications Workshop in honor of Rupert Lasser Helmholtz

More information

Interpolation via weighted l 1 minimization

Interpolation via weighted l 1 minimization Interpolation via weighted l 1 minimization Rachel Ward University of Texas at Austin December 12, 2014 Joint work with Holger Rauhut (Aachen University) Function interpolation Given a function f : D C

More information

List of Errata for the Book A Mathematical Introduction to Compressive Sensing

List of Errata for the Book A Mathematical Introduction to Compressive Sensing List of Errata for the Book A Mathematical Introduction to Compressive Sensing Simon Foucart and Holger Rauhut This list was last updated on September 22, 2018. If you see further errors, please send us

More information

On Optimal Frame Conditioners

On Optimal Frame Conditioners On Optimal Frame Conditioners Chae A. Clark Department of Mathematics University of Maryland, College Park Email: cclark18@math.umd.edu Kasso A. Okoudjou Department of Mathematics University of Maryland,

More information

Sparse solutions of underdetermined systems

Sparse solutions of underdetermined systems Sparse solutions of underdetermined systems I-Liang Chern September 22, 2016 1 / 16 Outline Sparsity and Compressibility: the concept for measuring sparsity and compressibility of data Minimum measurements

More information

On the l 1 -Norm Invariant Convex k-sparse Decomposition of Signals

On the l 1 -Norm Invariant Convex k-sparse Decomposition of Signals On the l 1 -Norm Invariant Convex -Sparse Decomposition of Signals arxiv:1305.6021v2 [cs.it] 11 Nov 2013 Guangwu Xu and Zhiqiang Xu Abstract Inspired by an interesting idea of Cai and Zhang, we formulate

More information

Guaranteed Minimum-Rank Solutions of Linear Matrix Equations via Nuclear Norm Minimization

Guaranteed Minimum-Rank Solutions of Linear Matrix Equations via Nuclear Norm Minimization Forty-Fifth Annual Allerton Conference Allerton House, UIUC, Illinois, USA September 26-28, 27 WeA3.2 Guaranteed Minimum-Rank Solutions of Linear Matrix Equations via Nuclear Norm Minimization Benjamin

More information

Compressed Sensing: Theory and Applications

Compressed Sensing: Theory and Applications Compressed Sensing: Theory and Applications Gitta Kutyniok March 12, 2012 Abstract Compressed sensing is a novel research area, which was introduced in 2006, and since then has already become a key concept

More information

c 2011 International Press Vol. 18, No. 1, pp , March DENNIS TREDE

c 2011 International Press Vol. 18, No. 1, pp , March DENNIS TREDE METHODS AND APPLICATIONS OF ANALYSIS. c 2011 International Press Vol. 18, No. 1, pp. 105 110, March 2011 007 EXACT SUPPORT RECOVERY FOR LINEAR INVERSE PROBLEMS WITH SPARSITY CONSTRAINTS DENNIS TREDE Abstract.

More information

Signal Recovery from Permuted Observations

Signal Recovery from Permuted Observations EE381V Course Project Signal Recovery from Permuted Observations 1 Problem Shanshan Wu (sw33323) May 8th, 2015 We start with the following problem: let s R n be an unknown n-dimensional real-valued signal,

More information

Sparse Recovery from Inaccurate Saturated Measurements

Sparse Recovery from Inaccurate Saturated Measurements Sparse Recovery from Inaccurate Saturated Measurements Simon Foucart and Jiangyuan Li Texas A&M University Abstract This article studies a variation of the standard compressive sensing problem, in which

More information

Z Algorithmic Superpower Randomization October 15th, Lecture 12

Z Algorithmic Superpower Randomization October 15th, Lecture 12 15.859-Z Algorithmic Superpower Randomization October 15th, 014 Lecture 1 Lecturer: Bernhard Haeupler Scribe: Goran Žužić Today s lecture is about finding sparse solutions to linear systems. The problem

More information

Elaine T. Hale, Wotao Yin, Yin Zhang

Elaine T. Hale, Wotao Yin, Yin Zhang , Wotao Yin, Yin Zhang Department of Computational and Applied Mathematics Rice University McMaster University, ICCOPT II-MOPTA 2007 August 13, 2007 1 with Noise 2 3 4 1 with Noise 2 3 4 1 with Noise 2

More information

6 Compressed Sensing and Sparse Recovery

6 Compressed Sensing and Sparse Recovery 6 Compressed Sensing and Sparse Recovery Most of us have noticed how saving an image in JPEG dramatically reduces the space it occupies in our hard drives as oppose to file types that save the pixel value

More information

Randomness-in-Structured Ensembles for Compressed Sensing of Images

Randomness-in-Structured Ensembles for Compressed Sensing of Images Randomness-in-Structured Ensembles for Compressed Sensing of Images Abdolreza Abdolhosseini Moghadam Dep. of Electrical and Computer Engineering Michigan State University Email: abdolhos@msu.edu Hayder

More information

Compressed sensing. Or: the equation Ax = b, revisited. Terence Tao. Mahler Lecture Series. University of California, Los Angeles

Compressed sensing. Or: the equation Ax = b, revisited. Terence Tao. Mahler Lecture Series. University of California, Los Angeles Or: the equation Ax = b, revisited University of California, Los Angeles Mahler Lecture Series Acquiring signals Many types of real-world signals (e.g. sound, images, video) can be viewed as an n-dimensional

More information