Expectation and Estimation
Multiple Random Variables
q 2차원 랜덤변수 사상S
S
Jx
y
1s
2s
2 1( ( ), ( ))
X s
Y s
1 2( ( ), ( ))
X s Y s
Y
X
Joint Cumulative Distribution Function
q Single Cumulative Distribution Function
q Joint Cumulative Distribution Function
Joint Cumulative Distribution Function
q Properties of Joint Cumulative Distribution Function
Joint Probability Density Function
q Joint Probability Density Function
Conditional Probability Density Function
q Statistical Independence
,
,
Conditional Probability Density Function
q Example , , 1 2 , 0 0 2 1 2 0 0Let the joint density of two random variables and be given by 1 ( 4 ), 0 2, 0 1 ( , ) 6 0, otherwise 1) ( , ) 0 and 1 ( , ) ( 4 ) 6 1 1 1 ( 4 ) 6 2 6 X Y X Y X Y x x X Y x y x y f x y f x y f x y dxdy x y dxdy x xy dy ¥ ¥ -¥ -¥ = = ì + < < < < ï = í ïî ³ = + é ù = ê + ú = ë û
ò ò
ò ò
ò
01 1 2 0 (2 8 ) 1 (2 4 ) 1 6Probability Density function
Expected Value
q Expected value or Mean of a discrete random variable X
- The expected value (or expectation) refers, intuitively, to the value of a random variable one would "expect" to find if one could repeat the random variable process an infinite number of times and take the
average of the values obtained.
- The expected value is a weighted average of all possible values.
Expected Value
q Expected value a random variable X(General meaning)
Expected Value
q Ex) Mean of normal pdf
Moments
q Expected value functions of a random variable X
q Moments about the origin
q Central Moments
[
( )]
( ) X ( ) E g X ¥ g x f x dx -¥ =ò
Let ( ) , 0,1,2,...then the n-th moment of is given by [ ] ( )
n n n n X g X X n X m E X ¥ x f x dx -¥ = = = =
ò
0 0 Let ( ) ( ) , 0,1,2,...where is the mean of , then the n-th central moment of
Moments
q Ex) 주사위를 1회 던졌을 때 나타나는 눈의 모멘트( )
( )
2 2 2 2 2 2 2 {1, 2, 3, 4, 5, 6}, ( ) 1/ 6, 1,2,...,6 1 1 21 7Mean (the 1 moment) : [ ] 1 ... 6
6 6 6 2
1 1 1 91
The 2 moment : [ ] 1 2 ... 6
6 6 6 6
Moments
q Ex) Gaussian Random Variable
2 2 ( ) 2 2 2 2 1 ( ) 2
Mean (the 1 moment) : [ ]
Variance(the 2 central moment) : [( ) ]
The pdf of the gaussian RV is represented by its mean and variance.
Moments
q Joint moments about the origin
q Correlation
,
For two random variables and ,
the joint moment is given by ( , ) ,
where is the order of the joint moment.
n k n k nk X Y X Y m E X Y x y f x y dxdy n k ¥ -¥ é ù = ë û = +
ò
[ ]
11 , ( , )If [ ] [ ] is satisfied, we say that there is
between and . Or, it is said that is statistically indpendent on no cor . relation XY X Y XY R m E XY x y f x y dxdy R E X E Y X Y X Y ¥ -¥ = = = =
ò
Moments
q Joint central moments
,
joint central mom
For two random variables and ,
-Moments
q Covariance 11 , , , , [ ] [ ] [ ] [ ] [ ] , 1 ( )( ) ( )( ) ( , ) ( , ) ( , ) ( , ) ( , ) XY X Y X Y X Y X Y E XY E XY XE Y XY E XY YE X XY X Y C E X X Y Y x X y Y f x y dxdyxyf x y dxdy X y f x y dxdy Y x f x y dxdy XY f x y dxdy
µ
¥ -¥ ¥ ¥ ¥ -¥ -¥ -¥ = = = = = = = ¥ -¥ = é ù = = ë - - û = - -= - -+ò
ò
ò
ò
ò
!"""#"""$ !"""#"""$ !"""#"""$ !""# no correlation or independent If 0( [ ] [ ]), there iseach other between and .
If and are orthogonal( 0), [ ] [ ] .
Multivariate random variables
q Multivariate random variables (Random vectors)
• Vectors with random variables
Multivariate random variables
q Mean and covariance under linear transformation
• Suppose the mean and the covariance of 𝑥 is given by
Mean: Covariance:
• A new variable via linear transformation 𝑦 = 𝐴𝑥. Then,
Multivariate Gaussian Random Variables
q Multivariate Gaussian Probability Density Function
Multivariate Gaussian Random Variables
q Conditioning
• Given a joint pdf 𝑓%'%$ 𝑥!, 𝑥" on R" • Measure 𝑥";
we would like to find the conditional pdf of𝑥!,
Multivariate Gaussian Random Variables
q Conditioning • Suppose 𝑥 ~ 𝑁 0, Σ with • Suppose we measure 𝑥" = 𝑦. • Conditional pdf 𝑥! given 𝑥" = 𝑦: 𝑥 = 𝑥𝑥! " Σ = Σ!! Σ!" Σ!"$ Σ"" 𝑓%' 𝑥! 𝑥" = 𝑦 = ! "- ) ,''&,'$, $$ &', '$ % 𝑒&'$ (&,'$,$$&'. % ,''&,'$,$$&','$% &' (&,'$,$$&'.
⟺ 𝑓%' 𝑥! 𝑥" = 𝑦 ~ 𝑁(Σ!"Σ""&!𝑦, Σ!! − Σ!"Σ""&!Σ!"$ )
Multivariate Gaussian Random Variables
q Estimation
• Suppose 𝑥 ~ 𝑁 0, Σ( , 𝑤 ~ 𝑁 0, Σ/ uncorrelated, and a measurement 𝑦 = 𝐴𝑥 + 𝑤 is given:
• We want the best estimate on 𝑥 given 𝑦, i.e., 𝐸[𝑥|𝑦] • Let 𝑧 = 𝑥𝑦 = 𝐼 0
𝐴 𝐼
𝑥 𝑤
• Then
• If Σ/ = 𝐼 and Σ( ⟶ ∞, this approaches to the least squares