MODULAR SYSTEMS & COMBINATORIAL OPTIMIZATION (based on course “System design”, 2004…2008,...
-
Upload
august-herbert-oliver -
Category
Documents
-
view
215 -
download
3
Transcript of MODULAR SYSTEMS & COMBINATORIAL OPTIMIZATION (based on course “System design”, 2004…2008,...
MODULAR SYSTEMS & COMBINATORIAL OPTIMIZATION
(based on course “System design”, 2004…2008, MIPT)
Mark Sh. LevinInst. for Inform. Transmission Problems, Russian Acad. of Sci.
Email: [email protected] Http://www.mslevin.iitp.ru/
Dept. ‘Technology for Complex System Modeling”,
Division of Applied Mathematics&Informatics, HSE, Moscow, Russia, Dec. 19, 2012
PLAN:1.Preliminaries (about me, about you)
2.Modularity (applications, basic technological problem)3.Decision cycle (problem, model, algorithm, computing, decisions)
4.Combinatorial optimization problems 5.Four-layer framework (basic combinatorial problems/models,
composite models, framework of problems, applied layer)6.Composite problems – research projects
7.Conclusion (about novelty)
PRELIMINARIES
About me: A. Education: (a) Radio Engineering (MTUSI, 1970) (b) Mehmat (MSU,1975)
(c) Faculty of Economics (PhD-studies, MSU)(1981) (d) PhD-engineering & CS (RAS)B. Works: (i) system design, (ii) software engineering: information software packages, (iii)management systems (iv)decision making + COMBINATORIAL OPTIMIZATION
Applications: (a) special systems, (b) governmental organizations (c) manufacturing, (d) geology, (e) house-building, (f) machine-building, (g) communication
C. Recent teaching: Moscow Univs.; MIPT (2004..2008): “System Design”
About you:1.Your future objectives
(i) Work in bank, consulting company, etc. (ii) Establishing new company (Google, Microsoft, Facebook, etc.)
(iii) Academic research & educational work:*BS level: 1...3 conf. papers
*MS level: 1…3 conf. papers, 1…3 journal articles (WoS)PhD level: 3…5 conf. papers, 3…5 journal articles (WoS)
Assistant Prof. level: 8…15 conf. papers, 5…7, journal articles (WoS) Associate Prof. level: 30 journal articles (WoS), 30 conf. papers
Full Prof. level: 60 journal articles (WoS), 60 conf. papers
GOALS: 1.Extending your thinking 2.Possible joint research 3.Usage of my materials
GLANCE
Applications (engineering, IT&CS, economics, geology, biomedicine, etc.)
System approach,
system design,systems
engineering
CombinatorialoptimizationHierarchical
systemmodel
MODULARITY
APPLICATIONS:(1)House-building (2)Computers (3)Machine-building (4)Bioinformatics (5)Software,
(6)Combinatorial chemistry, ETC
BASIC EXAMPLE: linguistics (NB!)
BASIC SYSTEM PROBLEMS: 1. System model (system hierarchy, architecture) 2. System design (system configuration) 3. System improvement/upgrade (adaptation, modification, reconfiguration) 4. Multi-stage system design (design of system trajectory) 5. Combinatorial system evolution (combinatorial modeling) 6. System forecasting
SYSTEM: 1.Basic (e.g., physical) system (e.g., computer, machine/car, house,
communication system, software, algorithm system, personnel) 2.Plan: medical treatment, plan of system improvement, economical plan, rules3.Requirements (system of requirements) 4.Standard (standard as system, system of standard)
COMBINATORIAL OPTIMIZATION PROBLEMS
COMBINATORIAL PROBLEMS (basic problems):There is a set of elements.
Find:Basic Problem 1. Ordering of elementsBasic Problem 2. Grouping of elements
Basic Problem 3. Assignment of elements to certain “places”
PROBLEMS:1.Ordering/scheduling 2.Ranking 3.Knapsack problem 4.Multiple choice problem
5.Assignemnt/allocation/location problems (including marriage problem)6.Clustering (grouping) 7.TSP 8.Graph coloring 9. Covering problems 10.Spanning trees (minimum spanning tree, Steiner tree) 11.SAT etc.
BASIC ISSUES:1.Complexity (Polynomial algorithm exists or does not exist)
2.Design of algorithms: (a) enumerative methods (B-A-B, dynamic programming) (b) polynomial algorithms (c) simple (e.g., greedy) heuristics (c) heuristics
(d) approximation algorithms (e) genetic algorithms (evolutionary optimization)
Multicriteria ranking/choice
Initialset ofalternatives
Linearranking
Choice Ranking
Clustering / Classification Problem
Initial setof objects/alternatives Clusters
Goals:1.To decrease the dimension2.To design a hierarchy
Simple structures (chains, trees, parallel-series graphs)
CHAIN
TREE
PARALLEL-SERIESGRAPH
Simple structures (hierarchy)
Level 4
Level 3
Level 1
Level 2
Optimization problems on graphs: illustrations
a0
a1
a2
a3
a4
a6
a5
a7
a8
a9
BASIC GRAPH (DIGRAPH):weights for arcs (or edges)
2
1
2
2
4
4
1
3 4
3
2
4 3
2
a0
a1
a2
a3
a4
a6
a5
a7
a8
a92
1
2
2
4
4
1
3 4
3
2
4 3
2
Shortest Path for < a0,a9 >:L = < a0,a1,a2,a3,a4,a7,a9 >2+1+1+2+2 = 8
Optimization problems on graphs: illustrations
a0
a1
a2
a3
a4
a6
a5
a7
a8
a9
Spanning tree (length = 19):
2
1
2
4
4
1
3 4
3
2
4 3
2
Traveling Salesman Problem :L = < a0,a1,a3,a5,a7,a9,a8,a4,a2,a6>2+1+3+4+2+2+3+4+4+4
a4
a3 a7
a1
a0 a2
a5 a6a9
a8
a0
a1
a2
a3
a4
a6
a5
a7
a8
a92
1
2
4
4
1
3 4
3
2
4 3
2
2
2
PACKING PROBLEM (illustration)
1
1
2
2
3
3
4
4
5
5
76
6
7
8
8
9
9
10
10
11
11
1213
14 . . .
REGION FORPACKING
ELEMENTS
GOALS:*Maximum of packed elements*Minimum offree space
BIN-PACKING PROBLEM (illustration)
CONTAINERS FORPACKING
ELEMENTS
. . .
1
1
2
2
3
3
4
4
5
5
6
6
GOAL:Usage of minimal number of containers
SCHEDULING: illustrative example for assembly process (algorithm of longest tails)
GOAL:Minimal totalcomplete time
1
3
2
6
5
4
9
8
7 10
11
12
13
14
15
16
17
18
19
Tasks &precedence constraints
1(distancefrom corner)
7
2
23
3
3
4
4
4
55
6
66
6 7
7
7
3 processors:
t
1
2
3
0
17
18
19
12
16
14
13
15
9
10
11
6
7
8
3
4
5 2 1
8
MAXIMA CLIQUE PROBLEM (illustration)
Initial graph G = (R, E), R is set of vertices, E is set of edges
Problem is: Find the maximal (by number of vertices) clique (i.e., complete subgraph)
G = (R,E)
Clique consisting of6 vertices(maximalcomplete subgraph)
Spanning (illustration): 1-connected graph
a0
a1
a2
a3
a4
a6
a5
a7
a8
a9
Steiner tree (example):
2
1
2
4
4
1
3 4
3
2
4 3
2
a4
a3 a7
a1
a0 a2
a5 a6
a9a8
2
a0
a1
a2
a3
a4
a6
a5
a7
a8
a9
Spanning tree (length = 19):
2
1
2
4
4
1
3 4
3
2
4 3
2
a4
a3 a7
a1
a0 a2
a5 a6a9
a8
2
Knapsack problem
max mi=1 ci xi
s.t. mi=1 ai xi b
xi {0, 1}, i = 1, … , m possible additional constraints m
i=1 aik xi bk , k = 1, … , l
. . . . . .
1 i m (index)
a1 ai am (required resource) c1 ci cm (utility / profit)
x1 xi xm (Boolean variable)
Multiple choice problem
max mi=1 qi
j=1 cij xij
s.t. mi=1 qi
j=1 aij xij b
qij=1 xij 1 , i = 1, … , m
xij {0, 1}, i = 1, … , m , j = 1, … , qi
. . . . . .
J1 Ji Jm
. . . . . .
. . .
i | Ji | = qi , j = 1, … , qi
Assignment/Allocation problem
Allocation (assignment, matching, location):
matrix of weights cij
BIPARTITE GRAPH
1
2
3
4
5
6
7
8
a
b
c
d
e
f
g
h
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
a b c d e f g h
1
2
3
4
5
6
7
8
Positions
Set of elements
Assignment/allocation problem
a3
a1
a2
an
b1
FORMULATION (algebraic):Set of elements: A = { a1 , … , ai , … , an }Set of positions: B = { b1 , … , bj , … . bm } (now let n = m)Effectiveness of pair ai and bj is: c ( ai , bj )
xij = 1 if ai is located into position bj and 0 otherwise ( xij { 0,1 } )
The problem is: max ni=1 n
j=1 cij xij
s.t. ni=1 xij = 1 j
nj=1 xij = 1 i
b2
b3
bm
. . . . . .
ELEMENTS POSITIONS
Multiple matching problem
A = { a1, … an } B = { b1, … bm }
C = { c1, … ck }
EXAMPLE:3-MATCHING(3-partitie graph)
Graph coloring problem (illustration)
Initial graph G = (A, E), A is set of vertices, E is set of edges
Problem is: Assign a color for each vertex with minimal number of colors under constraint: neighbor vertices have to have different colors
G = (A,E)
Right coloring
VERTEX COVERING PROBLEM
Vertex set A = { a1, … an }, edge set E={e1, …,ek}, graph G = (A, E)
PROBLEM: find vertex covering (A’ A) that covers e E
Satisfiability problem: illustration for application in software / electronic systems
SYSTEM
x1
xm
xm-1
x2
. . . y (0 or 1)
c1
c2
cn-1
cn
Example: c1 = not x1 OR x2 OR x4 OR not x5 OR x7
c2 = x1 OR not x2 OR not x3 OR x5 OR x7
c3 = not x1 OR not x2 OR x3 OR not x5 OR not xn
c4 = not x2 OR x3 OR x7 OR xn-2 OR xn-1
. . .
y = c1&c2& … &cn
Literal: xi / not xi
PROBLEM: Exist xo=(x1,…,xn) that y(xo) =1 OR not
Satisfiability Problem
SATISFIABILITY
3-SATISFIABILITY
3-MATCHING VERTEX COVERING
PARTITIONING (about knapsack)
HAMILTONIAN CYCLE CLIQUE
BASIC 6 NP-COMPLETE PROBLEMS AND DIAGRAM
Alignment (illustration)
CASE OF 2 WORDS:
A AB B D X
A DA C X Z
Word 1
Word 2
A AB B D X Z
Superstructure
C
A AB B D X
A DA C X Z
Morphological clique
PART 1
PART 2 PART 3
Vertices(design alternatives)
Edges(compatibility)
NOTE:about k-matching
COMPLEXITY OF COMBINATORIAL OPTIMIZAITON PROBLEMS
Polynomial solvableproblems
NP-hardproblems
Approximatepolynomial
solvableProblems(FPTAS)
Knapsackproblem
Multiple choiceproblem
Quadratic assignmentproblem
Morphologicalcliqueproblem
Cliqueproblem
TSP
DECISION CYCLE & Support Components
Applied modularproblem(s)
Solving process(e.g., computing)
DECISION
Frame of problems/mathematical models
Library of problems/ models
Solving schemesAlgorithms
Procedures
Program (software)/interactive procedure
Four-layer Framework
Multicriteriaranking
Knapsackproblem
Multiple choice
problem
Cluste-ring
Assignment/allocation
Design of hierarchy (clustering,
multicriteria spanning)
Morpholo-gical clique
(synthesis)
ETC
Layer 1: Basic problems/models
Multi-criteria
knapsack
Multicriteriamultiple choice
problem
Multi-criteria
assignment/allocation
ETC
Layer 2: Composite models/procedures
Four-stagecomposite
framework
Hierarchicalmorpho-logicaldesign
Systemupgrade/ improve-
ment
Systemevolution/
forecasting
ETC
Layer 3: Basic (typical) solving framework
Design/planning
oftesting
Modulardesign ofsoftware
Informa-tion
retrieval
Design ofmarke-
tingstrategy
ETC
Layer 4: Domain-oriented frameworks
Planningof
mainte-nance
Planningof
medicaltreatment
Improve-ment
ofnetwork
Evolutionof
require-ments
Span-ningtree
Steinertree
problem
Shortestpath
problem
Multistagedesign
S=X*Y*Z
Y Z=P*Q*U*V
Z1=P2*Q3*U1*V5
Z2=P1*Q2*U3*V1Y1
Y2
Y3
A B
A1
A2
A3
B1
B2
B3
B4
C1
C2
C3
C4
C5
D=I*J
I1
I2
I3
J1
J2
J3
J4
P1
P2
P3
Q1
Q2
Q3
Q4
U1
U2
U3
V1
V2
V3
V4
V5
V6
C
X=A*B*C*D
P Q U V
X1=A1*B2*C4*D3
X2=A3*B4*C2*D1
D1=I1*J1
D2=I1*J2
D3=I3*J4
S1=X2*Y3*Z2
S2=X1*Y2*Z1
JI
Illustration for HMMD approach
Compatibility for I,J
Compatibilityfor P,Q,U,V
Compatibilityfor A,B,C,D
Compatibilityfor X,Y,Z
Product trajectory
Stage 1
. . .
T0Stage 3
. . .
Stage 2
. . .
Trajectory
1
2
3
4
a
b
c
d
e
f
g
h
Groups of testers
SYSTEMS
FRAMEWORK: Clustering, Assignment, Multiple Choice Problem
TESTERS
2
Groupsof systems
X
X
XTest actions:A1 (no test)A2 (simple test)A3 (analysis test)A4 (structure research)A5 (new test)
MPEG 1
T0
MPEG 2
MPEG 4
Forecast
Improvement
Evolution as Generations of MPEG-like standard, Forecasting
SYSTEM LIFE CYCLE (product, etc.)
R & Dt
Manufacturing Testing MarketingUtilization &Maintenance Recycling
0 T
T: about 12 years (submarines, airplanes, nuclear technology, etc.) TENDENCY: increasing T (2 years, 6 months)
SYSTEM: 1.Basic (e.g., physical) system (e.g., computer, machine/car, house,
communication system, software, algorithm system, personnel) 2.Plan: medical treatment, plan of system improvement, economical plan, rules3.Requirements (system of requirements) 4.Standard (standard as system, system of standard)
Car:1.Body2.Motor3.Drive system4.Electronics5.Safety
MY COMBINATORIAL TECHNOLOGICAL SYSTEMS PROBLEMS:1.Design of system model2.System design3.System improvement4.System evolution5.System forecasting
Research Opportunity (with Novelty)
Novelty: new approach at each level/for each component:
1.Applied problem (engineering, CS, economics)
2.Model/Model framework
3.Algorithm / algorithm framework
4.Software (program, program package)
EXAMPLES OF STUDENT RESEARCH PROJECTS:ABOUT 30 PUBLISHED PAPERS (my site)
EXAMPLES OF MY RESEARCH PROJECTS:MY PUBLISHED PAPERS (my site)
That’s All
Thanks!http://www.mslevin.iitp.ru/
Mark Sh. Levin