text
stringlengths 0
89.3k
|
|---|
Θj |
ξΘjPKs |
i1Θi1jKs |
0 K s j24 |
We set the shape basis dimension Ksaccording to the as |
sumption of linear basis combination and truncate the sin |
gular values We can better measure the importance of dif |
ferent singular values by adjusting the weights through nor |
malization |
Solution for ˆSThe solution model for the optimization |
variable ˆScan be expressed as follows |
ˆS arg min |
ˆSβ |
2 |
SgˆSΛ |
2 |
FD |
Y1SgˆSΛE |
β |
2 |
ˆSQS |
2 |
FD |
Y2ˆSQSE |
25 |
where gis an invertible linear operator so we can rewrite |
the above equation in a more easily solvable form as |
ˆS arg min |
ˆSβ |
2 |
ˆSQS |
2 |
FD |
Y2ˆSQSE |
β |
2 |
g1SˆSΛ |
2 |
FD |
g1Y1 g1SˆSΛE |
26 |
The closedform solution of ˆScan be computed by taking |
the derivative of the model 26 and equating it to zero |
ˆSIPΛΛT |
QS1 |
βY2g1SΛT1 |
βg1Y1ΛT27 |
Solution for SFirst we define the firstorder smoothing |
matrix HR3F3Fas in 11 |
Hij |
1 j i i 1 3F1 |
1 ji 3 i 1 3F1 |
0Otherwise 28Algorithm 2 ADMM Optimization Algorithm |
1Input Initialize SSˆSSRpQΛ β β max λand |
ϵ 1e6 |
2repeat |
3 Update Sby Eq 22 yields S |
4 Update ˆSby Eq 27 yields ˆS |
5 Update Sby Eq 31 yields S |
6 Update Sby Eq 33 yields S |
7 repeat |
8 Calculate the gradient by Eq 35 and update Q |
by LM algorithm yields Q |
9 until Convergence |
10 Update Y1byY1β |
SgˆSΛ |
11 Update Y2byY2β |
ˆSQS |
12 Update Y3byY3β |
SRpS |
13 Update βasminβmax λβ |
14untilSS ϵ |
Output SQ |
By introducing the matrix H we can denote the TPA mod |
ule equivalently as |
F1X |
i1 |
QiSiQi1Si1 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.