rare event simulation using importance sampling and cross ... · a rare event problem denote x(t)...
TRANSCRIPT
![Page 1: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/1.jpg)
Rare Event Simulation usingImportance Sampling and
Cross-EntropyCaitlin James
Supervisor: Phil Pollett
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 1
![Page 2: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/2.jpg)
0 1 2 3 4 5 6 7 80
50
100
150
200
250
300
350
400
450
500Simulation of a population
Time t (years)
Pop
ulat
ion
size
X(t
)
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 2
![Page 3: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/3.jpg)
Outline
• A rare event problem• Stochastic SIS Logistic Epidemic (SIS)• Crude Monte Carlo Method (CMCM)• Importance Sampling (IS)• Cross-Entropy Method (CE)• Comparison of simulation estimates
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 3
![Page 4: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/4.jpg)
A rare event problem
Denote X(t) as the size of the population at time t.
Let N denote the maximum population size.
We wish to estimate
α = P(X(t) hits 0 before N).
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 4
![Page 5: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/5.jpg)
Notation• Suppose (X(t) : t ≥ 0) is a birth-death process
on finite space X = {0, 1, . . . , N}.
����
����
����
����
����
. . .0 1 2 N -1 N−→←− −→←− −→←− −→←−µ1 µ2 µN−1 µN
λ0 λ1 λN−2 λN−1
Figure 1: Transition diagram of a finite-state birth-death process
• Let (Xm,m = 0, 1, . . . ) denote the correspondingjump chain.
• Let A be the collection of all the sample paths of(Xm) & let A0 be the collection of all the pathsthat hit 0 before N .
• Let X = (X0, X1, . . . ) be a random sample pathof A and let A be the event {X ∈ A0}.
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 5
![Page 6: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/6.jpg)
Notation• Suppose (X(t) : t ≥ 0) is a birth-death process
on finite space X = {0, 1, . . . , N}.
• Let (Xm,m = 0, 1, . . . ) denote the correspondingjump chain.
• Let A be the collection of all the sample paths of(Xm) & let A0 be the collection of all the pathsthat hit 0 before N .
• Let X = (X0, X1, . . . ) be a random sample pathof A and let A be the event {X ∈ A0}.
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 5
![Page 7: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/7.jpg)
Notation• Suppose (X(t) : t ≥ 0) is a birth-death process
on finite space X = {0, 1, . . . , N}.
• Let (Xm,m = 0, 1, . . . ) denote the correspondingjump chain.
• Let A be the collection of all the sample paths of(Xm) & let A0 be the collection of all the pathsthat hit 0 before N .
• Let X = (X0, X1, . . . ) be a random sample pathof A and let A be the event {X ∈ A0}.
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 5
![Page 8: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/8.jpg)
Notation• Suppose (X(t) : t ≥ 0) is a birth-death process
on finite space X = {0, 1, . . . , N}.
• Let (Xm,m = 0, 1, . . . ) denote the correspondingjump chain.
• Let A be the collection of all the sample paths of(Xm) & let A0 be the collection of all the pathsthat hit 0 before N .
• Let X = (X0, X1, . . . ) be a random sample pathof A and let A be the event {X ∈ A0}.
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 5
![Page 9: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/9.jpg)
Equation
Then we can write
α = P(A)
= EfH(X) =
∫
A
H(x)f(x; u, P )µ(dx).
where• H(X) is the indicator function of the rare event
A defined by
H(x) =
{
1 if x ∈ A0
0 if x 6∈ A0.
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 6
![Page 10: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/10.jpg)
Equation
Then we can write
α = P(A)
= EfH(X) =
∫
A
H(x)f(x; u, P )µ(dx).
and• f(x; u, P ) = u(x0)
∏m−1k=0 P (xk, xk+1).
• u = (u(i) : i ∈ X ) is the initial distribution.• P = (P (i, j) : i, j ∈ X ) is the one-step transition
matrix of (Xm).
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 7
![Page 11: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/11.jpg)
Equation
Then we can write
α = P(A)
= EfH(X) =
∫
A
H(x)f(x; u, P )µ(dx).
and• f(x; u, P ) = u(x0)
∏m−1k=0 P (xk, xk+1).
• u = (u(i) : i ∈ X ) is the initial distribution.
• P = (P (i, j) : i, j ∈ X ) is the one-step transitionmatrix of (Xm).
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 7
![Page 12: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/12.jpg)
Equation
Then we can write
α = P(A)
= EfH(X) =
∫
A
H(x)f(x; u, P )µ(dx).
and• f(x; u, P ) = u(x0)
∏m−1k=0 P (xk, xk+1).
• u = (u(i) : i ∈ X ) is the initial distribution.• P = (P (i, j) : i, j ∈ X ) is the one-step transition
matrix of (Xm).
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 7
![Page 13: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/13.jpg)
Stochastic SIS LogisticEpidemic
• The SIS model is a finite-state birth and deathprocess for which the origin is an absorbingstate.
• The rate of infection per contact is denoted by λand µ denotes the per-capita death rate.
����
����
����
����
����
. . .0 1 2 N -1 N←− −→←− −→←− −→←−µ1 µ2 µN−1 µN
λ1 λN−2 λN−1
Figure 2: Transition diagram of SIS model
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 8
![Page 14: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/14.jpg)
Stochastic SIS LogisticEpidemic
• The SIS model is a finite-state birth and deathprocess for which the origin is an absorbingstate.
• The rate of infection per contact is denoted by λand µ denotes the per-capita death rate.
����
����
����
����
����
. . .0 1 2 N -1 N←− −→←− −→←− −→←−µ1 µ2 µN−1 µN
λ1 λN−2 λN−1
Figure 2: Transition diagram of SIS model
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 8
![Page 15: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/15.jpg)
Stochastic SIS LogisticEpidemic
The non-zero transition rates are defined as
λi = λ1
Ni(N − i) and µi = µi.
The jump chain (Xm) has jump probabilities
pi =λ(N − i)
µN + λ(N − i)
and qi = (1− pi) =µN
µN + λ(N − i).
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 9
![Page 16: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/16.jpg)
Stochastic SIS LogisticEpidemic
The non-zero transition rates are defined as
λi = λ1
Ni(N − i) and µi = µi.
The jump chain (Xm) has jump probabilities
pi =λ(N − i)
µN + λ(N − i)
and qi = (1− pi) =µN
µN + λ(N − i).
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 9
![Page 17: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/17.jpg)
Crude Monte Carlo Method
We can estimate our probability using CMCM. Thisinvolves simulating n replicates, X
1, . . . ,Xn, fromf( · ; u, P ) and setting
α =1
n
n∑
k=1
H(Xk).
We simulate our model until the process hits N or 0.We count 1 if the process hits 0.
The number of trials needed to get one successful trialhas a geometric distribution with the expected valuebeing 1/p, where p is the probability of a successfultrial.
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 10
![Page 18: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/18.jpg)
Crude Monte Carlo Method
We can estimate our probability using CMCM. Thisinvolves simulating n replicates, X
1, . . . ,Xn, fromf( · ; u, P ) and setting
α =1
n
n∑
k=1
H(Xk).
The number of trials needed to get one successful trialhas a geometric distribution with the expected valuebeing 1/p, where p is the probability of a successfultrial.
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 10
![Page 19: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/19.jpg)
CMCM example
If we start in State 8 and we wish to estimate theprobability of reaching 0 before 10, then we wouldrequire 1/α runs to see one successful path hit 0before 10. (Failure is hitting 10 before 0.)
The exact value of α starting in state 8 is 1.18E-05.Thus we need around 1/1.18E-05 ≈ 85, 000 runs tosee our first path hit 0 before 10.
If we use CMCM, then we would require many morethan 85, 000 runs to obtain a reasonable estimatefor α.
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 11
![Page 20: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/20.jpg)
CMCM example
If we start in State 8 and we wish to estimate theprobability of reaching 0 before 10, then we wouldrequire 1/α runs to see one successful path hit 0before 10. (Failure is hitting 10 before 0.)
The exact value of α starting in state 8 is 1.18E-05.Thus we need around 1/1.18E-05 ≈ 85, 000 runs tosee our first path hit 0 before 10.
If we use CMCM, then we would require many morethan 85, 000 runs to obtain a reasonable estimatefor α.
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 11
![Page 21: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/21.jpg)
CMCM example
If we start in State 8 and we wish to estimate theprobability of reaching 0 before 10, then we wouldrequire 1/α runs to see one successful path hit 0before 10. (Failure is hitting 10 before 0.)
The exact value of α starting in state 8 is 1.18E-05.Thus we need around 1/1.18E-05 ≈ 85, 000 runs tosee our first path hit 0 before 10.
If we use CMCM, then we would require many morethan 85, 000 runs to obtain a reasonable estimatefor α.
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 11
![Page 22: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/22.jpg)
Importance Sampling
• u to be an alternative initial distribution
• P to be an alternative transition matrix
• g(x; u, P ) = u(x0)∏m−1
k=0 P (xk, xk+1)
The following conditions must also hold:
u(i) > 0 implies u(i) > 0
and P (i, j) > 0 implies P (i, j) > 0.
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 12
![Page 23: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/23.jpg)
Importance Sampling
• u to be an alternative initial distribution
• P to be an alternative transition matrix
• g(x; u, P ) = u(x0)∏m−1
k=0 P (xk, xk+1)
The following conditions must also hold:
u(i) > 0 implies u(i) > 0
and P (i, j) > 0 implies P (i, j) > 0.
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 12
![Page 24: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/24.jpg)
Importance Sampling
• u to be an alternative initial distribution
• P to be an alternative transition matrix
• g(x; u, P ) = u(x0)∏m−1
k=0 P (xk, xk+1)
The following conditions must also hold:
u(i) > 0 implies u(i) > 0
and P (i, j) > 0 implies P (i, j) > 0.
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 12
![Page 25: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/25.jpg)
Importance Sampling
• u to be an alternative initial distribution
• P to be an alternative transition matrix
• g(x; u, P ) = u(x0)∏m−1
k=0 P (xk, xk+1)
The following conditions must also hold:
u(i) > 0 implies u(i) > 0
and P (i, j) > 0 implies P (i, j) > 0.
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 12
![Page 26: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/26.jpg)
Importance Sampling
Under the alternative measure g we can estimate α by
α = EfH(X) =
∫
A
H(x)f(x; u, P )
g(x; u, P )g(x; u, P )µ(dx)
= Eg
(
H(X)LT (X; u, P, u, P ); T <∞
)
,
where
Lm(x;u, P, u, P ) =f(x;u, P )
g(x; u, P )=
u(x0)∏m−1
k=0 P (xk, xk+1)
u(x0)∏m−1
k=0 P (xk, xk+1).
with T = inf{m : Xm = 0 or Xm = N} ("Stopping Time")
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 13
![Page 27: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/27.jpg)
Importance Sampling
Under the alternative measure g we can estimate α by
α = EfH(X) =
∫
A
H(x)f(x; u, P )
g(x; u, P )g(x; u, P )µ(dx)
= Eg
(
H(X)LT (X; u, P, u, P ); T <∞
)
,
where
Lm(x;u, P, u, P ) =f(x;u, P )
g(x; u, P )=
u(x0)∏m−1
k=0 P (xk, xk+1)
u(x0)∏m−1
k=0 P (xk, xk+1).
with T = inf{m : Xm = 0 or Xm = N} ("Stopping Time")
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 13
![Page 28: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/28.jpg)
Importance Sampling
Under the alternative measure g we can estimate α by
α = EfH(X) =
∫
A
H(x)f(x; u, P )
g(x; u, P )g(x; u, P )µ(dx)
= Eg
(
H(X)LT (X; u, P, u, P ); T <∞
)
,
where
Lm(x;u, P, u, P ) =f(x;u, P )
g(x; u, P )=
u(x0)∏m−1
k=0 P (xk, xk+1)
u(x0)∏m−1
k=0 P (xk, xk+1).
with T = inf{m : Xm = 0 or Xm = N} ("Stopping Time")
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 13
![Page 29: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/29.jpg)
Importance Sampling estimator
We can estimate α using the importance sampling(IS) estimator, defined by
α =1
n
n∑
i=1
H(X)Lm(X;u, P, u, P ),
where (recall)
Lm(x;u, P, u, P ) =f(x;u, P )
g(x; u, P )=
u(x0)∏m−1
k=0 P (xk, xk+1)
u(x0)∏m−1
k=0 P (xk, xk+1).
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 14
![Page 30: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/30.jpg)
Optimal IS density
Which change of measure gives the IS estimator withsmallest variance?
The smallest variance is obtained when g = g∗, theoptimal importance sampling density, given by
g∗(x) =|H(x)|f(x; u, P )
∫
A |H(x)|f(x; u, P )µ(dx).
If H(x) ≥ 0, then
g∗(x) =H(x)f(x; u, P )
α.
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 15
![Page 31: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/31.jpg)
Optimal IS density
Which change of measure gives the IS estimator withsmallest variance?
The smallest variance is obtained when g = g∗, theoptimal importance sampling density, given by
g∗(x) =|H(x)|f(x; u, P )
∫
A |H(x)|f(x; u, P )µ(dx).
If H(x) ≥ 0, then
g∗(x) =H(x)f(x; u, P )
α.
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 15
![Page 32: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/32.jpg)
Optimal IS density
Which change of measure gives the IS estimator withsmallest variance?
The smallest variance is obtained when g = g∗, theoptimal importance sampling density, given by
g∗(x) =|H(x)|f(x; u, P )
∫
A |H(x)|f(x; u, P )µ(dx).
If H(x) ≥ 0, then
g∗(x) =H(x)f(x; u, P )
α.
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 15
![Page 33: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/33.jpg)
Cross-Entropy Method
The Kullback-Leibler Cross-Entropy (CE) measuredefines a "distance" between two densities g and h
D(g, h) =
∫
g(x) lng(x)
h(x)µ(dx)
=
∫
g(x) ln g(x)µ(dx)−
∫
g(x) ln h(x)µ(dx).
The purpose of CE is to choose the IS density h suchthat the "distance" between the optimal IS density g∗
and density h is as small as possible.
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 16
![Page 34: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/34.jpg)
Properties of CE
1. D(· , ·) is non-symmetric ie: D(g, h) 6= D(h, g),thus D(g, h) is not a true distance between g andh in a formal sense, although
2. D(g, h) ≥ 0.
3. D(g, g) = 0.
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 17
![Page 35: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/35.jpg)
Restrict the IS density
If we restrict the density to belong to some family Fwhich contains the original density
f(x; u, P ) = u(x0)m−1∏
k=0
P (xk, xk+1)
and the alternative density
f(x; u, P ) = u(x0)m−1∏
k=0
P (xk, xk+1)
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 18
![Page 36: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/36.jpg)
CE optimisation problem
Then the CE method aims to solve the parametricoptimisation problem
min(u,P )
D(g∗, f( · ; u, P )),
where (recall)
g∗(x) =H(x)f( · ; u, P )
α.
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 19
![Page 37: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/37.jpg)
CE reference parameter
Since f( · ; u, P ) does not depend on (u, P ),minimising the CE distance between g∗ andf( · ; u, P ) is equivalent to maximising, with respectto (u, P ),∫
|H(x)|f(x;u, P ) ln f(x; u, P )µ(dx)
= E(u,P )|H(X)| ln f(X; u, P ).
Assuming H(X) ≥ 0, the optimal P (with respect toCE) is the solution to
P ∗ = arg maxP
E(u,P )H(X) ln f(X; u, P ).
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 20
![Page 38: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/38.jpg)
CE reference parameter
Since f( · ; u, P ) does not depend on (u, P ),minimising the CE distance between g∗ andf( · ; u, P ) is equivalent to maximising, with respectto (u, P ),∫
|H(x)|f(x;u, P ) ln f(x; u, P )µ(dx)
= E(u,P )|H(X)| ln f(X; u, P ).
Assuming H(X) ≥ 0, the optimal (u, P ) (withrespect to CE) is the solution to
(u∗, P ∗) = arg max(u,P )
E(u,P )H(X) ln f(X; u, P ).
Assuming H(X) ≥ 0, the optimal P (with respect toCE) is the solution to
P ∗ = arg maxP
E(u,P )H(X) ln f(X; u, P ).
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 20
![Page 39: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/39.jpg)
CE reference parameter
Since f( · ; u, P ) does not depend on (u, P ),minimising the CE distance between g∗ andf( · ; u, P ) is equivalent to maximising, with respectto (u, P ),∫
|H(x)|f(x;u, P ) ln f(x; u, P )µ(dx)
= E(u,P )|H(X)| ln f(X; u, P ).
Assuming H(X) ≥ 0, the optimal P (with respect toCE) is the solution to
P ∗ = arg maxP
E(u,P )H(X) ln f(X; u, P ).
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 20
![Page 40: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/40.jpg)
CE reference parameter
The optimal transition probability matrix is given by
P ∗(i, j) =E(u,P )H(X)
∑
k:Xk=i 1(Xk+1=j)
E(u,P )H(X)∑
k:Xk=i 1.
We can approximate this optimal transitionprobability matrix by implementing IS to obtain:
P ∗
l+1(i, j) ≈
∑
Xn
X=X1 H(X)Lm(X;u, P, u, Pl)
∑
k:Xk=i 1(Xk+1=j)∑
Xn
X=X1 H(X)Lm(X;u, P, u, Pl)
∑
k:Xk=i 1,
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 21
![Page 41: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/41.jpg)
CE reference parameter
The optimal transition probability matrix is given by
P ∗(i, j) =E(u,P )H(X)
∑
k:Xk=i 1(Xk+1=j)
E(u,P )H(X)∑
k:Xk=i 1.
We can approximate this optimal transitionprobability matrix by implementing IS to obtain:
P ∗
l+1(i, j) =E(u,Pl)
H(X)Lm(X;u, P, u, Pl)∑
k:Xk=i 1(Xk+1=j)
E(u,Pl)H(X)Lm(X;u, P, u, Pl)
∑
k:Xk=i 1
P ∗
l+1(i, j) ≈
∑
Xn
X=X1 H(X)Lm(X;u, P, u, Pl)
∑
k:Xk=i 1(Xk+1=j)∑
Xn
X=X1 H(X)Lm(X;u, P, u, Pl)
∑
k:Xk=i 1,
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 21
![Page 42: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/42.jpg)
CE reference parameter
The optimal transition probability matrix is given by
P ∗(i, j) =E(u,P )H(X)
∑
k:Xk=i 1(Xk+1=j)
E(u,P )H(X)∑
k:Xk=i 1.
We can approximate this optimal transitionprobability matrix by implementing IS to obtain:
P ∗
l+1(i, j) ≈
∑
Xn
X=X1 H(X)Lm(X;u, P, u, Pl)
∑
k:Xk=i 1(Xk+1=j)∑
Xn
X=X1 H(X)Lm(X;u, P, u, Pl)
∑
k:Xk=i 1,
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 21
![Page 43: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/43.jpg)
Initial CE parameter
Original parameters:
pi =λ(N − i)
µN + λ(N − i)and qi = (1− pi) =
µN
µN + λ(N − i).
Initial change of measures:
pi =µ(N − i)
λN + µ(N − i)and qi = (1− pi) =
λN
λN + µ(N − i).
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 22
![Page 44: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/44.jpg)
1 2 3 4 5 6 7 8 9 100
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1Estimation of the probability of reaching 0 before N
Initial state
Pro
babi
lity
exact CMC Importance Sampling estimation IS 95% CI
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 23
![Page 45: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/45.jpg)
1 2 3 4 5 6 7 8 9 100
0.005
0.01
0.015
0.02
0.025
0.03
0.035
0.04Estimation of the probability of reaching 0 before N
Initial state
Pro
babi
lity
exact CMC Importance Sampling estimation IS 95% CI
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 24
![Page 46: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/46.jpg)
Comparison of simulationestimates
N = 10, λ = 0.9, µ = 0.1, ρ = 0.11, X0 = 8,
No. of CE runs= 4
Sample Size Exact IS CMCM50 1.18E-05 1.40E-05 0
100 1.18E-05 1.64E-05 01000 1.18E-05 1.18E-05 02000 1.18E-05 1.18E-05 0
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 25
![Page 47: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/47.jpg)
Comparison of CE runsN = 10, λ = 0.9, µ = 0.1, ρ = 0.11, X0 = 8,
Sample Size = 1000, Exact = 1.18E-05
CE run IS 2 StD p1 p5 p9
0 1.47E-05 3.21E-06 0.091 0.053 0.0111 1.16E-05 1.79E-06 0.091 0.212 02 1.18E-05 4.12E-07 0.127 0.256 03 1.18E-05 1.38E-07 0.118 0.277 04 1.18E-05 9.09E-08 0.125 0.282 05 1.18E-05 5.89E-08 0.134 0.270 0
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 26
![Page 48: Rare Event Simulation using Importance Sampling and Cross ... · A rare event problem Denote X(t) as the size of the population at time t. Let N denote the maximum population size](https://reader033.vdocuments.us/reader033/viewer/2022042709/5f5236c426c49c096450dc99/html5/thumbnails/48.jpg)
Acknowledgements
• Phil Pollett• Joshua Ross• David Sirl• Ben Gladwin
Rare Event Simulation using Importance Sampling and Cross-Entropy – p. 27