Department of Computer Science

bankpottstownAI and Robotics

Oct 23, 2013 (3 years and 9 months ago)

80 views

Jun Han

John A. Miller

Department of Computer Science

University of Georgia


Gregory A Silver

College of
Business, Anderson
University


Introduction


Simulation
Optimization (SO)


Using SO for Glycomics


Overview of Glycomics


Glycan Quantification


Metabolic Pathways


Techniques for Simulation Optimization


SESSO Framework


Two Scenarios


Conclusions


Conceptual model


Domain Modeling


Simulation Designing
and Execution


Decision Parameter
Optimization

Optimization

Simulation

Model


History of Simulation Optimization from 1987


1987: “An art, not a science”


1998: Systematic
survey
and introduction





2000: A sub
-
chapter in simulation textbooks


Numerous application and research on how to
integrate optimization and simulation


2011: Regular track on Simulation Optimization in
WSC 2011

Decision parameter

Discrete

Continuous

Solution

Random Search

Gradient methods


Random search methods


Random walk, Simulated Annealing


Gradient based methods


Steepest descent, Conjugate gradient, BFGS


Heuristic methods


Genetic algorithm, Particle Swarm Optimization


Meta
-
modeling methods


Response surface methodology


Sample path optimization


Monte
Carlo Simulation


Glycan


produced by linking saccharides and attached to
proteins and lipids


Possible Applications


Cell differentiation


Disease processes


Cancer
Markers


Glycomics


“an integrated systems approach to structure
-
function relationships of glycans”


Identification


Quantification

Omics

Overview.

http://jdr.sagepub.com/citmgr?gca=spjdr;90/5/561

7

Experiments

Analysis


Label
-
free methods


Isotopic labeling


Static
IDAWG



Dynamic
IDAWG



Mass
Spectrometry


Modeling


Simulation


Optimization


Statistics


9


Metabolism


Biochemical reactions


Metabolic Network

GalNAc

(
mucin
-
type) core
synthesis/branching

http://www.ccrc.uga.edu/~moremen/glycomics/OglycanBranching/OglycanBranching/OglycanBranching.htm


SEESO
: A Semantically Enriched Environment
for
Simulation Optimization


Bootstrapped by


JSIM
: web
-
based simulation environment


ScalaTion
:
simulation
environment using domain
-
specific language (DSL)


DeMO
: Discrete
-
event Modeling Ontology


SoPT
: Simulation
oPTimization

ontology

Problem

Our Solution

Communication and sharing of
domain
model and
optimization problem

Ontology

Transformation from domain

Model to
optimization algorithm

Domain Specific
language (DSL)

Selection

of proper

optimization
algorithms

Rule
inferencing

min


[

(
𝒙
)
]

 



𝑉

𝒙




(
𝒙
)

0

𝒙





Use
Common
Random Number (CRN) to
reduce variance


Independent replications


Batch Means


Ranking and Selection


Simulator, Optimizer and (possible) Cost
Analyzer


Loosely Coupled


Iterative approach

Objective Function

Steepest Descent, etc.

m



(

)

 



10
>

>

0








def

solve (x0:
VectorD
):
VectorD

= {


var

x = x0
//
current point
var

xx
:
VectorD

= null
//
next point

var

gr:
VectorD

= null
//
gradient


breakable {



for
(k <
-

1 to MAX_ITER)
{



//
determine direction search


gr
= if (
usePartials
)
gradientD

(
df
, x)



// use
functions for partials



else gradient
(
fg
, x)



xx
=
lineSearch

(x,
gr)


if
(abs (
fg
(xx)
-

fg
(x)) < EPSILON) break


x
=
xx

}}
//
for


x

}
// solve


Establish connection between numerous real
world problems and optimization algorithms


Top level classes:


Optimization Component


Optimization Problem


Optimization Method

Conceptual Model

Experiment
design

DeMO

classes

Domain model

Decision
parameters

SoPT

classes

Optimization
Problem

Optimization
algorithm

Rule
Inferencing


DSL

Algorithm
selection,
configuration and
execution


A set of Rules


Rule
inferencing

(Rete algorithm)

if

(
ObjectiveFunction

is
quadratic_objective_function
) and
(
SolutionQuality

is
exact_solution
) and (
Constraint

is none) and
(
Restriction

is
real_restriction
)
then

(
OptimizationAlgorithm

is
Steepest_Descent
)


if

(
ObjectiveFunction

is

linear
_objective_function
) and (
SolutionQuality

is
exact_solution
) and (
Constraint

is
integer_constraint
) and (
Restriction

is
integer_restriction
)
then

(
OptimizationAlgorithm

is
Simplex_Algorithm
)


if

(
ObjectiveFunction

is

nonlinear
_objective_function
) and
(
SolutionQuality

is
heuristic_solution
) and (
Constraint

is none) and
(
Restriction

is
real_restriction
)
then

(
OptimizationAlgorithm

is
Genetic_Algorithm
)


Automatic Algorithm Configuration


Algorithm execution using DSL



Model definition using
DeMO


Code generation using
ScalaTion

DSL


Optimization algorithm selection using
SoPT


Optimization execution using DSL


Substrate (E), Product (P), Enzyme (E)




=

[
𝑃
]

=
𝑉
𝑚𝑎𝑥
[
𝑆
]
𝐾
𝑚
+
[
𝑆
]

ℎ 

𝐾
𝑚
=

𝑟
+

𝑐𝑎𝑡

𝑓


Decision Parameters


Rate constants


Temperatures,
Enzyme
concentration, gene expression
level, etc.

1

2

4

5

3

[

1
]
=

[
2

1
]
𝑋
2
𝑋
5

[

2
]
=

[
1

2
]
𝑋
1
𝑋
4
+

[
3

2
]
𝑋
3
[
𝑋
5
]

[

3
]
=

[
2

3
]
𝑋
2
[
𝑋
4
]


27


Quantitative glycomics
needs
simulation
optimization


Integration of ontology and DSL can facilitate
modeling, simulation and application of
simulation optimization for domain modelers