Wk 3-4 Genetic Programming Flashcards

1
Q

Genetic algorithms :

A

They are a specific type of evolutionary algorithm that that mimics the process of natural selection to solve optimization and search problems.

They employ genetic inspired operators (crossover, mutation) to evolve a population of individuals toward an optimal or near-optimal solution.

Focus representation and manipulation of genetic material (often in binary or chromosome format)

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
2
Q

Genetic algorithm outline ( same as EA) :

A
  • Generate random “programs”
  • Evaluate programs using training data
  • Selectively modify population of programs using crossover and mutation etc.
  • If a good program is found, finish, else go to
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
3
Q

How is the fitness evaluation different from standard EAs in genetic programs?

A

The chromosome is a program. To evaluate it, we have to run it, and test its behaviour over a range of potential inputs. Therefore its more complex.

EA Fitenss –> We just evaluate it and give it a score.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
4
Q

Genetic Programming Mutation:

A

Standard approach is to choose a subtree at random, remove it, and then generate a new subtree in its place.

This is usually biased towards nodes with high depth.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
5
Q

Five Major Preparatory Steps for GP

A
  • Determining the set of terminals
  • Determining the set of functions
  • Determining the fitness measure
  • Determining the parameters for the run
  • Determining the criterion for terminating a run
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
6
Q

In genetic programming how are functions represented ?

A

By parse trees

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
7
Q

Fitness parity problem

A

Input: A binary string of a fixed length.
Output: Determine if the binary string has an equal number of 0s and 1s.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
8
Q

(3-parity problem) odd-parity problem

A

That takes three input variables (X1, X2, X3) and returns 1 if an odd number of the input variables are 1. Otherwise, the function should return 0. ( toghether they make Y the function output I wish to achieve)

Serves as a common benchmark problem to evaluate the performance and effectiveness of GP algorithms.

You compare the program output with Y to see if the answer is correct.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
9
Q

Hamming distance

A

It calculates the minimum number of substitutions required to change one string into another by replacing individual characters. ( distance in changes between two strings of equal len )

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
10
Q

Lower Hamming distance

A

may be more likely to be chosen as parents for crossover, as they are more similar and can potentially generate offspring with desirable characteristics.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
11
Q

Higher Hamming

A

distance may be preferred for mutation to introduce diversity into the population.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
12
Q

Hamming distance crossover example

A

DIstance between parents is 3 .

Distance between cross over child and parent is 2 then the distance with the other parent will be 1.

H(PARENTA, CHILD) + H(PARENTB, CHILD) = H(A,B)

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
13
Q

SEMANTICS

A

Meaning of the problem, output fo the tree.

The semantics of a program is all input-output pairs defining the computed function

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
14
Q

SYNTAX

A

structure of the problem

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
15
Q

Traditional genetic programming

A

does blind syntactic manipulation of parent parse trees, regardless of their semantics

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
16
Q

Semantic Operators

A

Semantic operators act on the syntax but guarantee that some semantic criterion holds.

  • semantic mutation
  • semantic crossover
17
Q

semantic mutation

A

offspring semantically similar to parents.

example:
change to program tree but ensures the the ouput changes by only one position.

18
Q

semantic crossover

A

offspring semantically intermediate to parents

19
Q

Cone Landscape:

A

The theorem states that when GP uses semantic geometric operators, the fitness landscape becomes a cone landscape. In a cone landscape, fitness values uniformly decrease from a peak or optimum point in all directions. This means that the fitness landscape is smooth, has a single global optimum, and lacks local optima or plateaus.

20
Q

cone landscape advantages

A
  • Easy optimization in GP due to cone landscape
  • Quick convergence to the global optimum
  • Efficient exploration of the search space
  • Avoidance of local optima and deceptive landscapes