How to prove a matrix function is convex or nonconvex?
I have a function of three matrix variables. But now, the authors fix two of them, then update one, and I cannot understand how this function is convex in each iteration in the paper.
This formula is :
$$f(W,V,B) =|XW-V|^2_F +|Y-VB|^2_F +operatorname{tr}(V'LV) +2operatorname{tr}(W'DW),$$
where $X$, $Y$ are constant matrices and $L$ is constant laplace matrix. Suppose $D$ is a constant diagonal matrix.
Now, we fix two variables $W$ and $V$, then update $B$. How to solve?
If we do not fix any variables, how to explain that the objective function is non-convex?
In general, we use the Hessian matrix , but what should I do when the variable is a matrix?
convex-analysis convex-optimization machine-learning non-convex-optimization
asked Nov 28 at 7:11
learn_truth
12
add a comment |
I have a function of three matrix variables. But now, the authors fix two of them, then update one, and I cannot understand how this function is convex in each iteration in the paper.
This formula is :
$$f(W,V,B) =|XW-V|^2_F +|Y-VB|^2_F +operatorname{tr}(V'LV) +2operatorname{tr}(W'DW),$$
where $X$, $Y$ are constant matrices and $L$ is constant laplace matrix. Suppose $D$ is a constant diagonal matrix.
Now, we fix two variables $W$ and $V$, then update $B$. How to solve?
If we do not fix any variables, how to explain that the objective function is non-convex?
In general, we use the Hessian matrix , but what should I do when the variable is a matrix?
convex-analysis convex-optimization machine-learning non-convex-optimization
asked Nov 28 at 7:11
learn_truth
12
Welcome to MSE. For some basic information about writing mathematics at this site see, e.g., basic help on mathjax notation, mathjax tutorial and quick reference, main meta site math tutorial and equation editing how-to.
– José Carlos Santos
Nov 28 at 7:17
B appears in only one place. So all you need to know is that the square of Frobenius norm is convex.
– Michal Adamaszek
Nov 28 at 9:29
@Michal Adamaszek If the function is non-convex, we can use the anti-evidence method. But when only B is a matrix variable, the objective function is a convex function. Is it appropriate to use this anti-evidence method: (y-vb)^ 2?
– learn_truth
Nov 28 at 12:44
@lilili My hint was not about proving general non-convexity. It was about how to prove convexity in B: Frobenius norm square is convex and the rest is linear.
– Michal Adamaszek
Nov 28 at 13:24
@Michal Adamaszek The variable is a matrix, does the rule still apply?
– learn_truth
Nov 29 at 3:09
add a comment |
I have a function of three matrix variables. But now, the authors fix two of them, then update one, and I cannot understand how this function is convex in each iteration in the paper.
This formula is :
$$f(W,V,B) =|XW-V|^2_F +|Y-VB|^2_F +operatorname{tr}(V'LV) +2operatorname{tr}(W'DW),$$
where $X$, $Y$ are constant matrices and $L$ is constant laplace matrix. Suppose $D$ is a constant diagonal matrix.
Now, we fix two variables $W$ and $V$, then update $B$. How to solve?
If we do not fix any variables, how to explain that the objective function is non-convex?
In general, we use the Hessian matrix , but what should I do when the variable is a matrix?
convex-analysis convex-optimization machine-learning non-convex-optimization
asked Nov 28 at 7:11
learn_truth
12
I have a function of three matrix variables. But now, the authors fix two of them, then update one, and I cannot understand how this function is convex in each iteration in the paper.
This formula is :
$$f(W,V,B) =|XW-V|^2_F +|Y-VB|^2_F +operatorname{tr}(V'LV) +2operatorname{tr}(W'DW),$$
where $X$, $Y$ are constant matrices and $L$ is constant laplace matrix. Suppose $D$ is a constant diagonal matrix.
Now, we fix two variables $W$ and $V$, then update $B$. How to solve?
If we do not fix any variables, how to explain that the objective function is non-convex?
In general, we use the Hessian matrix , but what should I do when the variable is a matrix?
convex-analysis convex-optimization machine-learning non-convex-optimization
convex-analysis convex-optimization machine-learning non-convex-optimization
asked Nov 28 at 7:11
learn_truth
12
asked Nov 28 at 7:11
learn_truth
12
edited Dec 9 at 12:57
asked Nov 28 at 7:11
learn_truth
12
asked Nov 28 at 7:11
learn_truth
12
asked Nov 28 at 7:11
learn_truth
12
12
Welcome to MSE. For some basic information about writing mathematics at this site see, e.g., basic help on mathjax notation, mathjax tutorial and quick reference, main meta site math tutorial and equation editing how-to.
– José Carlos Santos
Nov 28 at 7:17
B appears in only one place. So all you need to know is that the square of Frobenius norm is convex.
– Michal Adamaszek
Nov 28 at 9:29
@Michal Adamaszek If the function is non-convex, we can use the anti-evidence method. But when only B is a matrix variable, the objective function is a convex function. Is it appropriate to use this anti-evidence method: (y-vb)^ 2?
– learn_truth
Nov 28 at 12:44
@lilili My hint was not about proving general non-convexity. It was about how to prove convexity in B: Frobenius norm square is convex and the rest is linear.
– Michal Adamaszek
Nov 28 at 13:24
@Michal Adamaszek The variable is a matrix, does the rule still apply?
– learn_truth
Nov 29 at 3:09
add a comment |
Welcome to MSE. For some basic information about writing mathematics at this site see, e.g., basic help on mathjax notation, mathjax tutorial and quick reference, main meta site math tutorial and equation editing how-to.
– José Carlos Santos
Nov 28 at 7:17
B appears in only one place. So all you need to know is that the square of Frobenius norm is convex.
– Michal Adamaszek
Nov 28 at 9:29
@Michal Adamaszek If the function is non-convex, we can use the anti-evidence method. But when only B is a matrix variable, the objective function is a convex function. Is it appropriate to use this anti-evidence method: (y-vb)^ 2?
– learn_truth
Nov 28 at 12:44
@lilili My hint was not about proving general non-convexity. It was about how to prove convexity in B: Frobenius norm square is convex and the rest is linear.
– Michal Adamaszek
Nov 28 at 13:24
@Michal Adamaszek The variable is a matrix, does the rule still apply?
– learn_truth
Nov 29 at 3:09
Welcome to MSE. For some basic information about writing mathematics at this site see, e.g., basic help on mathjax notation, mathjax tutorial and quick reference, main meta site math tutorial and equation editing how-to.
– José Carlos Santos
Nov 28 at 7:17
Welcome to MSE. For some basic information about writing mathematics at this site see, e.g., basic help on mathjax notation, mathjax tutorial and quick reference, main meta site math tutorial and equation editing how-to.
– José Carlos Santos
Nov 28 at 7:17
B appears in only one place. So all you need to know is that the square of Frobenius norm is convex.
– Michal Adamaszek
Nov 28 at 9:29
B appears in only one place. So all you need to know is that the square of Frobenius norm is convex.
– Michal Adamaszek
Nov 28 at 9:29
@Michal Adamaszek If the function is non-convex, we can use the anti-evidence method. But when only B is a matrix variable, the objective function is a convex function. Is it appropriate to use this anti-evidence method: (y-vb)^ 2?
– learn_truth
Nov 28 at 12:44
@Michal Adamaszek If the function is non-convex, we can use the anti-evidence method. But when only B is a matrix variable, the objective function is a convex function. Is it appropriate to use this anti-evidence method: (y-vb)^ 2?
– learn_truth
Nov 28 at 12:44
@lilili My hint was not about proving general non-convexity. It was about how to prove convexity in B: Frobenius norm square is convex and the rest is linear.
– Michal Adamaszek
Nov 28 at 13:24
@lilili My hint was not about proving general non-convexity. It was about how to prove convexity in B: Frobenius norm square is convex and the rest is linear.
– Michal Adamaszek
Nov 28 at 13:24
@Michal Adamaszek The variable is a matrix, does the rule still apply?
– learn_truth
Nov 29 at 3:09
@Michal Adamaszek The variable is a matrix, does the rule still apply?
– learn_truth
Nov 29 at 3:09
add a comment |
1 Answer
1
active
oldest
votes
I want to give an answer to the first question, expanding what was already hinted. The goal is to prove that the function $B mapsto f(V,W,B) $ is convex for every $V,W$ fixed (and lying in some space which is not specified by the OP). Since $B$ appears in the second term only, we investigate it alone: $$ | Y - VB |_F ^2 = langle Y - VB , Y - VB rangle_F = |Y|_F ^2 + |V B |_F^2 - 2 langle Y , VB rangle_F. $$ For $B_1$ and $B_2$ matrices we get $$ begin{split} | Y - V (tB_1 + (1-t)B_2) |_F ^2 & =|Y|_F ^2 + |V (tB_1 + (1-t)B_2) |_F^2 - 2 langle Y , V(tB_1 + (1-t)B_2) rangle_F \ & le |Y|_F ^2 + t | VB_1 |_F ^2 + (1-t) |VB_2 |_F ^2 \ & - 2( t langle Y ,VB_1rangle_F + (1-t)langle Y ,VB_2rangle _F) end{split} $$where I have used the (bi)linearity of the inner product and the elementary fact that if $f,g $ are two convex functions with $g$ nondecreasing then $g circ f$ is a convex function; in our specific case $ g(x) = x^2$ and $f (X) = |X|_F$ (norms are convex).
answered Nov 30 at 16:52
gangrene
915514
norm(V(tB1 + (1-t)B2, 'fro')^2 <= t*norm(VB1,'fro')^2 + (1-t)*norm(VB2,'fro')^2 ,why? This does not conform to the triangle inequality.
– learn_truth
Dec 7 at 2:42
It is justified below, norms are convex and composition between nondecreasing convex and convex gives you another convex function.
– gangrene
Dec 7 at 13:10
add a comment |
Your Answer
StackExchange.ifUsing("editor", function () {
return StackExchange.using("mathjaxEditing", function () {
StackExchange.MarkdownEditor.creationCallbacks.add(function (editor, postfix) {
StackExchange.mathjaxEditing.prepareWmdForMathJax(editor, postfix, [["$", "$"], ["\\(","\\)"]]);
});
});
}, "mathjax-editing");
StackExchange.ready(function() {
var channelOptions = {
tags: "".split(" "),
id: "69"
};
initTagRenderer("".split(" "), "".split(" "), channelOptions);
StackExchange.using("externalEditor", function() {
// Have to fire editor after snippets, if snippets enabled
if (StackExchange.settings.snippets.snippetsEnabled) {
StackExchange.using("snippets", function() {
createEditor();
});
}
else {
createEditor();
}
});
function createEditor() {
StackExchange.prepareEditor({
heartbeatType: 'answer',
autoActivateHeartbeat: false,
convertImagesToLinks: true,
noModals: true,
showLowRepImageUploadWarning: true,
reputationToPostImages: 10,
bindNavPrevention: true,
postfix: "",
imageUploader: {
brandingHtml: "Powered by u003ca class="icon-imgur-white" href="https://imgur.com/"u003eu003c/au003e",
contentPolicyHtml: "User contributions licensed under u003ca href="https://creativecommons.org/licenses/by-sa/3.0/"u003ecc by-sa 3.0 with attribution requiredu003c/au003e u003ca href="https://stackoverflow.com/legal/content-policy"u003e(content policy)u003c/au003e",
allowUrls: true
},
noCode: true, onDemand: true,
discardSelector: ".discard-answer"
,immediatelyShowMarkdownHelp:true
});
}
});
Sign up or log in
StackExchange.ready(function () {
StackExchange.helpers.onClickDraftSave('#login-link');
});
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Post as a guest
Required, but never shown
StackExchange.ready(
function () {
StackExchange.openid.initPostLogin('.new-post-login', 'https%3a%2f%2fmath.stackexchange.com%2fquestions%2f3016847%2fhow-to-prove-a-matrix-function-is-convex-or-nonconvex%23new-answer', 'question_page');
}
);
Post as a guest
Required, but never shown
1 Answer
1
active
oldest
votes
1 Answer
1
active
oldest
votes
active
oldest
votes
active
oldest
votes
I want to give an answer to the first question, expanding what was already hinted. The goal is to prove that the function $B mapsto f(V,W,B) $ is convex for every $V,W$ fixed (and lying in some space which is not specified by the OP). Since $B$ appears in the second term only, we investigate it alone: $$ | Y - VB |_F ^2 = langle Y - VB , Y - VB rangle_F = |Y|_F ^2 + |V B |_F^2 - 2 langle Y , VB rangle_F. $$ For $B_1$ and $B_2$ matrices we get $$ begin{split} | Y - V (tB_1 + (1-t)B_2) |_F ^2 & =|Y|_F ^2 + |V (tB_1 + (1-t)B_2) |_F^2 - 2 langle Y , V(tB_1 + (1-t)B_2) rangle_F \ & le |Y|_F ^2 + t | VB_1 |_F ^2 + (1-t) |VB_2 |_F ^2 \ & - 2( t langle Y ,VB_1rangle_F + (1-t)langle Y ,VB_2rangle _F) end{split} $$where I have used the (bi)linearity of the inner product and the elementary fact that if $f,g $ are two convex functions with $g$ nondecreasing then $g circ f$ is a convex function; in our specific case $ g(x) = x^2$ and $f (X) = |X|_F$ (norms are convex).
answered Nov 30 at 16:52
gangrene
915514
norm(V(tB1 + (1-t)B2, 'fro')^2 <= t*norm(VB1,'fro')^2 + (1-t)*norm(VB2,'fro')^2 ,why? This does not conform to the triangle inequality.
– learn_truth
Dec 7 at 2:42
It is justified below, norms are convex and composition between nondecreasing convex and convex gives you another convex function.
– gangrene
Dec 7 at 13:10
add a comment |
I want to give an answer to the first question, expanding what was already hinted. The goal is to prove that the function $B mapsto f(V,W,B) $ is convex for every $V,W$ fixed (and lying in some space which is not specified by the OP). Since $B$ appears in the second term only, we investigate it alone: $$ | Y - VB |_F ^2 = langle Y - VB , Y - VB rangle_F = |Y|_F ^2 + |V B |_F^2 - 2 langle Y , VB rangle_F. $$ For $B_1$ and $B_2$ matrices we get $$ begin{split} | Y - V (tB_1 + (1-t)B_2) |_F ^2 & =|Y|_F ^2 + |V (tB_1 + (1-t)B_2) |_F^2 - 2 langle Y , V(tB_1 + (1-t)B_2) rangle_F \ & le |Y|_F ^2 + t | VB_1 |_F ^2 + (1-t) |VB_2 |_F ^2 \ & - 2( t langle Y ,VB_1rangle_F + (1-t)langle Y ,VB_2rangle _F) end{split} $$where I have used the (bi)linearity of the inner product and the elementary fact that if $f,g $ are two convex functions with $g$ nondecreasing then $g circ f$ is a convex function; in our specific case $ g(x) = x^2$ and $f (X) = |X|_F$ (norms are convex).
answered Nov 30 at 16:52
gangrene
915514
norm(V(tB1 + (1-t)B2, 'fro')^2 <= t*norm(VB1,'fro')^2 + (1-t)*norm(VB2,'fro')^2 ,why? This does not conform to the triangle inequality.
– learn_truth
Dec 7 at 2:42
It is justified below, norms are convex and composition between nondecreasing convex and convex gives you another convex function.
– gangrene
Dec 7 at 13:10
add a comment |
I want to give an answer to the first question, expanding what was already hinted. The goal is to prove that the function $B mapsto f(V,W,B) $ is convex for every $V,W$ fixed (and lying in some space which is not specified by the OP). Since $B$ appears in the second term only, we investigate it alone: $$ | Y - VB |_F ^2 = langle Y - VB , Y - VB rangle_F = |Y|_F ^2 + |V B |_F^2 - 2 langle Y , VB rangle_F. $$ For $B_1$ and $B_2$ matrices we get $$ begin{split} | Y - V (tB_1 + (1-t)B_2) |_F ^2 & =|Y|_F ^2 + |V (tB_1 + (1-t)B_2) |_F^2 - 2 langle Y , V(tB_1 + (1-t)B_2) rangle_F \ & le |Y|_F ^2 + t | VB_1 |_F ^2 + (1-t) |VB_2 |_F ^2 \ & - 2( t langle Y ,VB_1rangle_F + (1-t)langle Y ,VB_2rangle _F) end{split} $$where I have used the (bi)linearity of the inner product and the elementary fact that if $f,g $ are two convex functions with $g$ nondecreasing then $g circ f$ is a convex function; in our specific case $ g(x) = x^2$ and $f (X) = |X|_F$ (norms are convex).
answered Nov 30 at 16:52
gangrene
915514
I want to give an answer to the first question, expanding what was already hinted. The goal is to prove that the function $B mapsto f(V,W,B) $ is convex for every $V,W$ fixed (and lying in some space which is not specified by the OP). Since $B$ appears in the second term only, we investigate it alone: $$ | Y - VB |_F ^2 = langle Y - VB , Y - VB rangle_F = |Y|_F ^2 + |V B |_F^2 - 2 langle Y , VB rangle_F. $$ For $B_1$ and $B_2$ matrices we get $$ begin{split} | Y - V (tB_1 + (1-t)B_2) |_F ^2 & =|Y|_F ^2 + |V (tB_1 + (1-t)B_2) |_F^2 - 2 langle Y , V(tB_1 + (1-t)B_2) rangle_F \ & le |Y|_F ^2 + t | VB_1 |_F ^2 + (1-t) |VB_2 |_F ^2 \ & - 2( t langle Y ,VB_1rangle_F + (1-t)langle Y ,VB_2rangle _F) end{split} $$where I have used the (bi)linearity of the inner product and the elementary fact that if $f,g $ are two convex functions with $g$ nondecreasing then $g circ f$ is a convex function; in our specific case $ g(x) = x^2$ and $f (X) = |X|_F$ (norms are convex).
answered Nov 30 at 16:52
gangrene
915514
answered Nov 30 at 16:52
gangrene
915514
answered Nov 30 at 16:52
gangrene
915514
answered Nov 30 at 16:52
gangrene
915514
915514
norm(V(tB1 + (1-t)B2, 'fro')^2 <= t*norm(VB1,'fro')^2 + (1-t)*norm(VB2,'fro')^2 ,why? This does not conform to the triangle inequality.
– learn_truth
Dec 7 at 2:42
It is justified below, norms are convex and composition between nondecreasing convex and convex gives you another convex function.
– gangrene
Dec 7 at 13:10
add a comment |
norm(V(tB1 + (1-t)B2, 'fro')^2 <= t*norm(VB1,'fro')^2 + (1-t)*norm(VB2,'fro')^2 ,why? This does not conform to the triangle inequality.
– learn_truth
Dec 7 at 2:42
It is justified below, norms are convex and composition between nondecreasing convex and convex gives you another convex function.
– gangrene
Dec 7 at 13:10
norm(V(tB1 + (1-t)B2, 'fro')^2 <= t*norm(VB1,'fro')^2 + (1-t)*norm(VB2,'fro')^2 ,why? This does not conform to the triangle inequality.
– learn_truth
Dec 7 at 2:42
norm(V(tB1 + (1-t)B2, 'fro')^2 <= t*norm(VB1,'fro')^2 + (1-t)*norm(VB2,'fro')^2 ,why? This does not conform to the triangle inequality.
– learn_truth
Dec 7 at 2:42
It is justified below, norms are convex and composition between nondecreasing convex and convex gives you another convex function.
– gangrene
Dec 7 at 13:10
It is justified below, norms are convex and composition between nondecreasing convex and convex gives you another convex function.
– gangrene
Dec 7 at 13:10
add a comment |
Thanks for contributing an answer to Mathematics Stack Exchange!
- Please be sure to answer the question. Provide details and share your research!
But avoid …
- Asking for help, clarification, or responding to other answers.
- Making statements based on opinion; back them up with references or personal experience.
Use MathJax to format equations. MathJax reference.
To learn more, see our tips on writing great answers.
Some of your past answers have not been well-received, and you're in danger of being blocked from answering.
Please pay close attention to the following guidance:
- Please be sure to answer the question. Provide details and share your research!
But avoid …
- Asking for help, clarification, or responding to other answers.
- Making statements based on opinion; back them up with references or personal experience.
To learn more, see our tips on writing great answers.
Sign up or log in
StackExchange.ready(function () {
StackExchange.helpers.onClickDraftSave('#login-link');
});
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Post as a guest
Required, but never shown
StackExchange.ready(
function () {
StackExchange.openid.initPostLogin('.new-post-login', 'https%3a%2f%2fmath.stackexchange.com%2fquestions%2f3016847%2fhow-to-prove-a-matrix-function-is-convex-or-nonconvex%23new-answer', 'question_page');
}
);
Post as a guest
Required, but never shown
Sign up or log in
StackExchange.ready(function () {
StackExchange.helpers.onClickDraftSave('#login-link');
});
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Post as a guest
Required, but never shown
Sign up or log in
StackExchange.ready(function () {
StackExchange.helpers.onClickDraftSave('#login-link');
});
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Post as a guest
Required, but never shown
Sign up or log in
StackExchange.ready(function () {
StackExchange.helpers.onClickDraftSave('#login-link');
});
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Post as a guest
Required, but never shown
Required, but never shown
Required, but never shown
Required, but never shown
Required, but never shown
Required, but never shown
Required, but never shown
Required, but never shown
Required, but never shown
Welcome to MSE. For some basic information about writing mathematics at this site see, e.g., basic help on mathjax notation, mathjax tutorial and quick reference, main meta site math tutorial and equation editing how-to.
– José Carlos Santos
Nov 28 at 7:17
B appears in only one place. So all you need to know is that the square of Frobenius norm is convex.
– Michal Adamaszek
Nov 28 at 9:29
@Michal Adamaszek If the function is non-convex, we can use the anti-evidence method. But when only B is a matrix variable, the objective function is a convex function. Is it appropriate to use this anti-evidence method: (y-vb)^ 2?
– learn_truth
Nov 28 at 12:44
@lilili My hint was not about proving general non-convexity. It was about how to prove convexity in B: Frobenius norm square is convex and the rest is linear.
– Michal Adamaszek
Nov 28 at 13:24
@Michal Adamaszek The variable is a matrix, does the rule still apply?
– learn_truth
Nov 29 at 3:09