Slide 31
Slide 31 text
Recap: With Dual Gradient Descent
2019/7/17
31
( ) ( )
,
min . .
t t
l s t
=
u x
( ) ( ) ( )
1 1
1 1
,... , ,... ,
1
min , . . ,
T T
T
t t t t t t t
t
c s t f
− −
=
= =
u u x x
x u x x u u x
( ) ( )
( )
( ) ( )
( )2
1 1
, ,
T T
t t t t t t
t t
L l
= =
= +
− + −
x u x u
( ) ( ) ( )
( )
* *
, ,
g L
=
( )
* argmin , ,
L
=
( )
* *
* *
* *
, ,
dg dL d dL d dL
d d dg d dg d
= + +
( )
* argmin , ,
L
=
Lを最小にした なので勾配は0
(Appendix:全微分の公式,参照)
,
1. Find
2. Find
3.
*
dg
d
+
( )
* argmin , ,
L
=
*
( )
* argmin , ,
L
=
軌道最適化(iLQR等で解法)
教師あり学習(SGD等で解法)