User:Molokaicat: Difference between revisions

From Cornell University Computational Optimization Open Textbook - Optimization Wiki
Jump to navigation Jump to search
Tags: Replaced Visual edit
 
Line 1: Line 1:
MEng Systems Engineering student
MEng Systems Engineering student          
MS and BS in Statistics
 
== Introduction ==
                                                                                                                                                                                                                                                                                                                             
 
Introduction
 
x ∈ R N often not really sparse but approximately sparse
 
Φ ∈ R M × N for M � N is a Random Gaussian or Bernoulli matrix
 
y ∈ R M are the observed y samples
 
e ∈ R M noise vector k e k 2 ≤ η
 
s.t.
 
How can we reconstruct x from y = Φx + e
 
The goal is to reconstruct x 0 ∈ R N given y and Φ
 
Sensing matrix Φ must satisfy RIP i.e. Random Gaussian or Bernoulli ma-
 
traceries satisfies (cite)
 
let [ N ] = { 1, . . . , N } be an index set
 
[ N ] enumerates the columns of Φ and x
 
Φ is an undetermined systems with infinite solutions. Why l 2 norm does
 
not work
 
What is compression is sunomunus with to the sparsity.
 
The problem formulation is to recover sparse data x 0 ∈ R N
 
The support of x 0 is supp ( x 0 ) = { i ∈ [ N ] : x 0 ( i ) 6 = 0 }
 
we say x is k sparse when | supp ( x )| ≤ k
 
We are interested in the smallest supp ( x ) , i.e. min(supp ( x ) )
 
Restricted isometry property (RIP) Introduced by Candes, Tao
 
Random Gaussian and Bernoulli satisfies RIP
 
Let Φ ∈ R M × N satsify RIP, Let [ N ] be an index set
 
Before we get into RIP lets talk about RIC
 
Restricted Isometry Constant (RIC) is the smallest δ | s s.t. s ⊆ [ N ] that satsifies the RIP condition
 
For s is a restriction on x denoted by x | s x ∈ R N tos k-sparse x s.t. RIP is
 
satisfied the s = | Γ | i.e. s ⊆ [ N ] and Φ | s ⊆ Φ where the columns of Φ | s is
 
indexed by i ∈ S
 
 
x i , i f i ∈ S
 
( x | S ) i =
 
0 otherwise
 
RIP defined as
 
( 1 − δ s )k x k 22 ≤ k Φx k 22 ≤ ( 1 + δ s )k x k 22
 
3 Lemmas Page 267 Blumensath Davies IHT for CS
 
Lemma 1(Blumensath, Davis 2009 Iterative hard thresholding for compressed
 
sensing), For all index sets Γ and all Φ for which RIP holds with s = | Γ | that is
 
s = supp ( x )
 
1k Φ Γ T k 2 ≤
 
q
 
1 + δ | Γ | k y k 2
 
( 1 − δ | Γ | )k x Γ k 22 ≤ k Φ Γ T Φ Γ x Γ k 22 ≤ ( 1 + δ | Γ | )k x Γ k 22
 
and
 
k( I − Φ Γ T Φ Γ )k 2 ≤ δ | Γ | k x Γ k 2
 
SupposeΓ ∩ Λ = ∅
 
k Φ Γ T Φ Λ ) x Λ k 2 ≤ δ s k x Λ k 2
 
Lemma 2 (Needell Tropp, Prop 3.5 in CoSaMP: Iterative signal recovery
 
from incomplete and inaccurate √ samples)
 
If Φ satisfies RIP k Φx s k 2 ≤ 1 + δ s k x s k 2 , ∀ x s : k x s k 0 ≤ s, Then ∀ x
 
k Φx k 2 ≤
 
p
 
1 + δ s k x k 2 +
 
p
 
1 + δ s
 
k x k 1
 
sqrts
 
Lemma 3 (Needell Tropp, Prop 3.5 in CoSaMP: Iterative signal recovery
 
from incomplete and inaccurate samples)
 
Let x s be the best s-term approximation to x. Let x r = x − x s Let
 
y = Φx + e = Φx s + Φx r + e = Φx s + ẽ
 
If Φ statisfies RIP for sparcity s, then the norm of error ẽ is bounded by
 
k ẽ k 2 ≤
 
p
 
1 + δ s k x − x s k 2 +
 
p
 
1 + δ s
 
k x − x s k 1
 
 
+ k e k 2
 
s
 
∀ x
 
Theory
 
Gel’fend n-width
 
Errors E ( S, Φ, D )
 
Definition Mutual Coherence
 
LetA ∈ R M × N , themutualcoherenceµ A isde f inedby :
 
µ A =
 
|h a i , a j i|
 
k a i kk a j k
 
i 6 = j
 
We want a small µ A because it will be close ot the normal matrix, which
 
will satisfie RIP
 
Numerical Example
 
Basis Persuit
 
Iterative Hard Thresholding IHT
 
Solver
 
2
<math>(1 - \delta_s) \| x \|_2 ^2 \leq \| \Phi x \|_2^2 \leq (1 + \delta_s) \| x \|_2 ^2</math>        
 
 
                               

Latest revision as of 01:22, 16 December 2021

MEng Systems Engineering student