BABEŞ-BOLYAI UNIVERSITY

Faculty of Computer Science and Mathematics

ARTIFICIAL
INTELLIGENCE

Intelligent systems
Machine learning
Genetic Programming
 Topics
A. Short introduction in Artificial Intelligence (AI)

A. Solving search problems

A. Definition of search problems
B. Search strategies
A. Uninformed search strategies
B. Informed search strategies
C. Local search strategies (Hill Climbing, Simulated Annealing, Tabu Search, Evolutionary
algorithms, PSO, ACO)
D. Adversarial search strategies

C. Intelligent systems
A. Rule-based systems in certain environments
B. Rule-based systems in uncertain environments (Bayes, Fuzzy)
C. Learning systems
A. Decision Trees
B. Artificial Neural Networks
C. Evolutionary algorithms
D. Support Vector Machines
D. Hybrid systems

May, 2017 Artificial Intelligence - Intelligent systems (GP) 2

 Useful information
 Chapter 15 of C. Groşan, A. Abraham, Intelligent Systems: A
Modern Approach, Springer, 2011

 Chapter 9 of T. M. Mitchell, Machine Learning, McGraw-Hill

Science, 1997

 Documents from 12_svm and 13_GP folders

May, 2017 Artificial Intelligence - Intelligent systems (GP) 3

 Intelligent systems

Expert
systems Decision trees
Bayes
Fuzzy Artificial Neural Networks
Support Vector Machines
Rule-based systems
Evolutionary Algorithms

Objects,
frames,
agents

Knowledge based systems (KBS) Computational intelligence

May, 2017 Artificial Intelligence - Intelligent systems (GP) 4

 Intelligent systems – Machine Learning
 Typology

 Experience criteria:
 Supervised learning
 Unsupervised learning
 Active learning
 Reinforcement learning

 Algorithm criteria
 Decision trees
 Artificial Neural Networks
 Evolutionary Algorithms
 Support Vector Machines
 Hidden Markov Models

May, 2017 Artificial Intelligence - Intelligent systems (GP) 5

 Intelligent systems – machine learning
 Genetic programming (GP)
 Definition
 Design
 Advantages
 Limits
 Versions

May, 2017 Artificial Intelligence - Intelligent systems (GP) 6

 Intelligent systems - GP
Remember

 Supervised learning  regression problem (study of relations among variables)

 For a set of n data (examples, instances, cases)
 Training data – as pairs ((atribute_datai, outputi), where
 i =1,n (n = # of training data)
 atribute_datai= (atri1, atri2, ..., atrim), m – # of attributes (characteristics, properties) of an example
 outputi – a real number
 Testing data
 (atribute_datai), i =n+1,N (N-n = # of testing data)
 Determine
 An (unknown) function that maps the attributes info outputs on training data
 Output (value) of a (new) test data by using the learnt function (on training data)

 How we find the function (expression)?

 Evolutionary algorithms  genetic programming

May, 2017 Artificial Intelligence - Intelligent systems (GP) 7

 Intelligent systems - GP
Remember
 Evolutionary algorithms
 Nature-inspired (bio-inspired)
 Iterative
 Based on
 Populations of potential solutions
 Random search guided by
 Natural selection operation
 Crossover and mutation operations
 Parallel processing of more solutions
 Evolutionary metaphor
Natural evolution Problem solving

Individual Possible solution

Population Set of possible solutions

Chromosome Solution coding (representation)

Gene Part of representation

Fitness (adaptation Quality

measure)

Crossover and mutation Search operators

Environment Problem search space

May, 2017 Artificial Intelligence - Intelligent systems (GP)

 Intelligent systems - GP
Remember

 Evolutionary algorithm

Initialisation P(0)
Crossover
Evaluation P(0)
g := 0; //generation
while (not stop_condition) execute

Recombination
Repeat

selection

Mutation
Select 2 parents p1 and p2 from P(g)
Crossover(p1,p2)  o1 and o2
Mutation(o1)  o1*
Mutation(o2)  o2*
Evaluation(o1*) Survival
Evaluation(o2*) selection
Add o1* and o2* in P(g+1)
until P(g+1) is complete
g := g + 1 Population Population
EndWhile (parents) (offspring)

May, 2017 Artificial Intelligence - Intelligent systems (GP) 9

 Intelligent systems - GP
 Definition
 Proposed by Koza in 1988
 http://www.genetic-programming.org/
 A special case of evolutionary algorithms
 Chromosome
 As trees that encode small programs
 Fitness of a chromosome
 Performance of the program encoded by the
chromosome

 GP’s aim
 Evolving computer programs
 Gas evolve solutions for particular problems only
May, 2017 Artificial Intelligence - Intelligent systems (GP) 10
 Intelligent systems - GP
 Design
 Chromosome representation
 Very important, but a difficult task
 Chromosome = tree with nodes of type
 Function  (mathematical) operators
(+,-,*,/,sin,log, if,...)
 Terminal  attributes of data or constants
(x,y,z,a,b,c,...)
 that encodes the mathematical expression of the
program (regression problem  function expression)

x(y+2) x+y2+3
May, 2017 Artificial Intelligence - Intelligent systems (GP) 11
 Intelligent systems - GP
 Design
 Fitness
 Prediction error – difference between what we want to obtain and
what we actually obtain

 For a regression problem with input data (2 attributes and an

output) and 2 chromosomes:
 c1 = 3x1-x2+5
 c2 = 3x1+2x2+2 f*(x1,x2)=3x1+2x2+1 – unknown

x1 x2 f*(x1,x2) f1(x1,x2) f2(x1,x2) |f*-f1| |f*-f2|

1 1 6 7 7 1 1
0 1 3 4 4 1 1
1 0 4 8 5 4 1
-1 1 0 1 1 1 1
c is better
∑=7 ∑= 4 2

than c1
May, 2017 Artificial Intelligence - Intelligent systems (GP)
 Intelligent systems - GP
 Design
 Fitness
 Prediction error – difference between what we want to obtain and
what we actually obtain

 For a classification problem with input data (2 attributes and an

output) and 2 chromosomes:
 c1 = 3x1-x2+5
 c2 = 3x1+2x2+2

x1 x2 f*(x1,x2) f1(x1,x2) f2(x1,x2) |f*-f1| |f*-f2|

1 1 Yes Yes Yes 0 0
0 1 No Yes No 1 0
1 0 Yes No No 1 1
-1 1 Yes No yes 1 0
c is better
∑=3 ∑= 1 2

than c1
May, 2017 Artificial Intelligence - Intelligent systems (GP)
 Intelligent systems - GP
 Design
 Chromosome initialisation
 Random generation of correct trees  valid programs (valid
mathematical expressions)
 Establish a maximal depth of the trees Dmax
 3 initialisation methods
 Full  each branch of the root has depth Dmax
 Nodes of depth d < Dmax are initialised by a function from F
 Nodes of depth d = Dmax are initialised by a terminal from T
 Grow  each branch of the root has a depth < Dmax
 Nodes of depth d < Dmax are initialised by an element from F  T
 Nodes of depth d = Dmax are initialised by a terminal from T
 Ramped half and half  ½ of population is initialised by using full
method and ½ of population is initialised by grow method

May, 2017 Artificial Intelligence - Intelligent systems (GP) 14

 Intelligent systems - GP
 Design
 Genetic operators  recombination selection
 Similar to other EAs
 Advise  proportional selection

 over-selection  for very large populations

 Sort the population based on fitness and consider 2 groups:
 Group 1 contains the best x% chromosomes from population
 Group 2 contains (100-x)% chromosome from population
 For populations with 1000, 2000, 4000 or 8000
chromosomes, x can be 32%, 16%, 8% and 4% respectively
 80% of selection operators will choose chromosomes from the first group
and 20% of them from the second group

May, 2017 Artificial Intelligence - Intelligent systems (GP) 15

 Intelligent systems - GP
 Design
 Genetic operators  survival selection
 Sketches
 Generational
 Steady-state
 Problems
 Bloat  the fattest individual survives (size of chromosomes
increases during evolution)
 Solutions
 Block the variation operators that produce to fat offsprings
 parsimony pressure -  to give a penalty in the cost function
(or fitness function) to long programs or program with many
non-coding parts

May, 2017 Artificial Intelligence - Intelligent systems (GP) 16

 Intelligent systems - GP
 Design
 Genetic operators  crossover and mutation
 Parameters
 A probability p of choosing between XO and mutation
 p = 0 (cf. Koza) or p = 0.05 (cf. Banzhaf)
 Two probabilities pc and pm for establishing the part
of chromosome(s) that must be changed
 Size of offsprings can be different to size of parents

May, 2017 Artificial Intelligence - Intelligent systems (GP) 17

 Intelligent systems - GP
 Design
 Genetic operators  crossover
 By cutting point
 sub-trees are exchanged
 Cutting point is randomly generated

p1=(x+y)*(z-sin(x)) f1=(x+y)yz

p2=xyz+x2

f2=(z-sin(x))x+x2

May, 2017 Artificial Intelligence - Intelligent systems (GP) 18

 Intelligent systems - GP
 Design
 Genetic operators  mutation
 Grow mutation  replace a leaf by a new sub-tree

p=(x+y)*(z-sin(x)) f=(x+y)*(z-sin(x+4))

May, 2017 Artificial Intelligence - Intelligent systems (GP) 19

 Intelligent systems - GP
 Design
 Genetic operators  mutation
 Shrink mutation  replace a sub-tree by a leaf

p=(x+y)*(z-sin(x)) f=(x+y)*4

May, 2017 Artificial Intelligence - Intelligent systems (GP) 20

 Intelligent systems - GP
 Design
 Genetic operators  mutation
 Koza mutation  replace a node (leaf or internal) by a
sub-tree
p=(x+y)*(z-sin(x)) f=(x+y)*sin(x+4)

May, 2017 Artificial Intelligence - Intelligent systems (GP) 21

 Intelligent systems - GP
 Design
 Genetic operators  mutation
 Switch mutation
 Select an internal node and re-order its sub-trees

 Cycle mutation
 Select a node and replace it by a new node of the
same type (internal node with a function, leaf node
with a terminal)

May, 2017 Artificial Intelligence - Intelligent systems (GP) 22

 Intelligent systems - GP
 GAs vs. GP
 Chromosome’s shape
 GAs – linear chromosomes
 GP – non-linear chromosomes
 Chromosome’s size
 GAs – fix size
 GP – variable size (depth or width)

 Offspring generation
 GAs – XO and mutation
 GP – XO or mutation

May, 2017 Artificial Intelligence - Intelligent systems (GP) 23

 Intelligent systems - GP
 Advantages
 GP finds solutions for problems without an optimal
solution
 A program for car driving  there are more solution
 Some solutions  safe but slow driving
 Other solutions  dangerous, but fast driving
 Car driving  trade-off large speed and safety
 GP is useful for problems whose variables are frequently
changed
 Car driving on a highway
 Car driving on a forest road

 Limits
 Large time required for evolving the solution
May, 2017 Artificial Intelligence - Intelligent systems (GP) 24
 Intelligent systems - GP
 GP versions
 Linear GP (Cramer, Nordin)
 Gene Expression Programming (Ferreira)
 Multi Expression Programing (Oltean)
 Gramatical Evolution (Ryan, O’Neill)
 Cartesian Genetic Programming (Miller)

May, 2017 Artificial Intelligence - Intelligent systems (GP) 25

 Intelligent systems - GP
 Linear GP
 Evolving programs written in an imperative language (fitness computation
does not require interpretation)  works fast
 Representation
 Vector of statements, each statement being (in the case of a maximal arrity of n
for a function)
 Index_op, out_register, in1_register, in2_register,…, inn_register

May, 2017 Artificial Intelligence - Intelligent systems (GP) 26

 Intelligent systems - GP
 Linear GP

 Initialisation
 Random
 Constraints
 Initial length of chromosome (# of statements)

 Variation genetic operators

 Crossover – 2-cutting points
 Mutation
 Micro-mutation  change an operand or operator (without
modifying the size of chromosome)
 Macro-mutation  insert or eliminate a statement (modifying the
size of chromosome)

May, 2017 Artificial Intelligence - Intelligent systems (GP) 27

 Intelligent systems - GP
 Linear GP

 Advantages
 Evolution into a low-level language

 Disadvantages
 # of required registers (# of problem’s attributes)

 Resources
 Register Machine Learning Technologies http://www.aimlearning.com
 Peter Nordin's home page http://fy.chalmers.se/~pnordin
 Wolfgang Banzhaf's home page http://www.cs.mun.ca/~banzhaf
 Markus Brameier's home page http://www.daimi.au.dk/~brameier

May, 2017 Artificial Intelligence - Intelligent systems (GP) 28

 Intelligent systems - GP
 Gene Expression Programming (GEP)
 Main idea
 Linear representation of expressions that can be encoded in a tree (by
breadth-first traversing procedure)

 Representation
 A chromosome is composed by more genes
 Linked by + or *
 Each gene is composed by
 Head
 Contains functions and terminals
 Tail
 Contains t terminals only, where t = (n-1)*h+1, with n –
maximal arrity of a function from F

May, 2017 Artificial Intelligence - Intelligent systems (GP) 29

 Intelligent systems - GP
 GEP
 Initialisation
 Randomly, by elements from F and T (cf. to previous rules)

 Variation genetic operators

 Crossover
 At allele level
 One cutting point
 Two cutting points
 At gene level
 Chromosomes exchange (between them) some genes (located on corresponding
positions)
 Mutation
 At allele level
 Change an element from head or tail (following the previous initialisation rules)
 Transpositions
  the introduction of an insertion sequence somewhere in a chromosome

May, 2017 Artificial Intelligence - Intelligent systems (GP) 30

 Intelligent systems - GP
 GEP
 Advantages
 Coding into chromosomes some correct programs due to gene splitting in head and tail

 Disadvantages
 Multi-gene chromosomes
 How many genes?
 How to link the genes?

 Resources
 Gene Expression Programming website, http://www.gepsoft.com
 Heitor Lopes's home page
http://www.cpgei.cefetpr.br/~hslopes/index-english.html
 Xin Li's home page http://www.cs.uic.edu/~xli1
 GEP in C# http://www.c-sharpcorner.com/Code/2002/Nov/GEPAlgorithm.asp

May, 2017 Artificial Intelligence - Intelligent systems (GP) 31

 Intelligent systems - GP
 Multi Expression Programming (MEP)
 Main idea
 Chromosome is composed by more genes, each gen being a 3 address
code
 Similarly to GEP, but faster

 Representation
 Linear
 A gene contains a (binary or unary) function and
pointers to its arguments
 Chromosome encodes more possible solutions  each
solution corresponds to a gene
 Quality of a solution (gene) = sum (over training
data) of differences between what we want to obtain
and what we obtain
 Fitness = quality of the best gene

May, 2017 Artificial Intelligence - Intelligent systems (GP) 32

 Intelligent systems - GP
 MEP
 Initialisation
 First gene must be a terminal
 Other genes can contain
 A terminal or
 A (unary or binary) function and pointers to its arguments
 Arguments of a function located in the i th gene must be located in
genes of index < I

 Variation genetic operators

 Crossover  exchange some genes between parents

 1-cutting point
 2-cutting points
 Uniform
 Mutation  modify a gene

 First gene  randomly generate a new terminal

 Other genes  randomly generate a terminal or a function (function symbol
and its arguments)

May, 2017 Artificial Intelligence - Intelligent systems (GP) 33

 Intelligent systems - GP
 MEP
 Advantages
 Dynamic output for each chromosome
 Complexity of the search program (expression)
 Programs (expression) of variable length obtained without special
operators
 Program of exponential length encoded into chromosomes of polynomial
length

 Disadvantages
 Complexity of decoding for unknown training data  evolving game’s
strategies

 Resources
 Mihai Oltean's home page http://www.cs.ubbcluj.ro/~
 Crina Gro»san's home page http://www.cs.ubbcluj.ro/~cgrosan
 MEP web page http://www.mep.cs.ubbcluj.ro
 MEP in C# http://www.c-sharpcorner.com

May, 2017 Artificial Intelligence - Intelligent systems (GP) 34

 Intelligent systems - GP
 Grammatical Evolution (GE)
 Main idea
 Evolving programs in Backus-Naur form (program expressed as a grammar with terminal symbols, non-
terminals, start symbol and rules)
 Representation
 Binary strings of codons (groups of 8 bits)  rule that must be applied
 Example
 G={N,T,S,P}, N={+, -,*, /, sin, (, )}, T= {expr, op2, op1}, S=expr, P is:
 expr ::=a|b|c| expr op2 expr|(expr op2 expr)| op1 expr
 op2::=+|-|*|/,
 op1::=sin
 C*GE=(9 12 12 3 15 7 11 4 2 5 0 6 11 0 1 7 12)
 S= expr  expr op2 expr  aop2 expr a + expr  a + expr op2 expr
 a + expr op2 expr op2 expr  a + b op2 expr op2 expr

May, 2017 Artificial Intelligence - Intelligent systems (GP) 35

 Intelligent systems - GP
 GE
 Initialisation
 Binary string is randomly initialised by 0 or 1 (without constraints)  valid programs
 Decoding ends when a complete program is obtained
 If the codons end and the program is incomplete, restart the codons from beginning 
wrapping

 Variation genetic operators

 Crossover
 Cutting point XO
 Mutation
 Probabilistic change of a bit into its opponent
 Duplication
 A sequence of genes is copied to the end of chromosome
 Pruning
 Elimination of unused genes

May, 2017 Artificial Intelligence - Intelligent systems (GP) 36

 Intelligent systems - GP
 GE
 Advantages
 Evolving programs written in languages whose statements can be expressed as BNF rules
 Representation can be changed by changing the grammar

 Disadvantages
 Infinite wrapping  limit the repetitions and penalise the chromosomes that overpass a given
threshold of repetitions

 Resources
 Grammatical Evolution web page, http://www.grammatical-evolution.org
 Conor Ryan's home page, http://www.csis.ul.ie/staff/conorryan
 Michael O'Neill's home page, http://ncra.ucd.ie/members/oneillm.html
 John James Collins's home page, http://www.csis.ul.ie/staff/jjcollins
 Maarten Keijzer's home page, http://www.cs.vu.nl/~mkeijzer
 Anthony Brabazon's home page http://ncra.ucd.ie/members/brabazont.html

May, 2017 Artificial Intelligence - Intelligent systems (GP) 37

 Intelligent systems - GP
 Cartesian Genetic Programming (CGP)
 Main idea
 Chromosomes as graphs (matrix)  more complex programs

 Representation
 Cartesian system (matrix of nodes)
 A node has associated
 A function
 Inputs
 Outputs

 Chromosome output
 Output of any node

May, 2017 Artificial Intelligence - Intelligent systems (GP) 38

 Intelligent systems - GP
 CGP
 Initialisation
 Randomly
 Inputs of any node must be nodes from previous columns
 Nodes of the first column has as inputs the problem attributes

 Variation genetic operators

 Crossover
 Is not applied
 Mutation
 Modify the elements of a node

May, 2017 Artificial Intelligence - Intelligent systems (GP) 39

 Intelligent systems - GP
 CGP
 Advantages
 Evolving the index of node that provides the output of the program
encoded into the chromosome
 Evolved program can have one or more outputs

 Disadvantages
 number of columns influences the results

 Resources
 Julian. F. Miller's home page http://www.elec.york.ac.uk/intsys/users/j
fm7
 Lukás Sekanina's home page http://www.fit.vutbr.cz/~sekanina/

May, 2017 Artificial Intelligence - Intelligent systems (GP) 40

 Intelligent systems - GP
 Applications
 Problems with relations between inputs and
outputs

 Regression problems

 Classification problems

May, 2017 Artificial Intelligence - Intelligent systems (GP) 41

 Intelligent systems - GP
 Applications
 Design problems

 Evolving digital circuits

 Evolving antenna
 http://idesign.ucsc.edu/projects/evo_antenna.html

 Evolving programs

 Evolving pictures and music

 http://www.cs.vu.nl/~gusz/

 Others
 http://www.genetic-programming.com/humancompetitive.

May, 2017 Artificial Intelligence - Intelligent systems (GP) 42

 Review
 Machine learning
 Genetic programming (GP)
 Evolutionary algorithms with chromosomes as trees
 Chromosomes
 Trees
 Matrix
 Linear
 Encode potential solutions
 Mathematical expressions  regression/classification
problems
 Boolean expressions  Even Parity problems or digital
circuits design
 Programs  evolving source codes for problem solving
May, 2017 Artificial Intelligence - Intelligent systems (GP) 43
 Next lecture
A. Short introduction in Artificial Intelligence (AI)

A. Solving search problems

A. Definition of search problems
B. Search strategies
A. Uninformed search strategies
B. Informed search strategies
C. Local search strategies (Hill Climbing, Simulated Annealing, Tabu Search, Evolutionary
algorithms, PSO, ACO)
D. Adversarial search strategies

C. Intelligent systems
A. Rule-based systems in certain environments
B. Rule-based systems in uncertain environments (Bayes, Fuzzy)
C. Learning systems
A. Decision Trees
B. Artificial Neural Networks
C. Evolutionary algorithms
D. Support Vector Machines
D. Hybrid systems

May, 2017 Artificial Intelligence - Intelligent systems (GP) 44

 Next lecture – useful information

 Chapter 15 of C. Groşan, A. Abraham, Intelligent Systems: A

Modern Approach, Springer, 2011

 Chapter 9 of T. M. Mitchell, Machine Learning, McGraw-Hill

Science, 1997

 Documents from svm folder

May, 2017 Artificial Intelligence - Intelligent systems (GP) 45

  Presented information have been inspired from different bibliographic
sources, but also from past AI lectures taught by:

 PhD. Assoc. Prof. Mihai Oltean – www.cs.ubbcluj.ro/~moltean

 PhD. Assoc. Prof. Crina Groşan - www.cs.ubbcluj.ro/~cgrosan

 PhD. Prof. Horia F. Pop - www.cs.ubbcluj.ro/~hfpop

May, 2017 Artificial Intelligence - Intelligent systems (GP) 46