Chapter 6

Question 1

genetic algorithms: typically applied to?

Answer 1

genetic algos: applied to

1. discrete function optimization
2. as a benchmark
3. to straightforward problems with binary representaiton

Question 2

evolution strategies: what do the chromosomes look like

Answer 2

ES chromosomes: 3 parts

1. object vars

strategy params:

2. mutation step size
3. rotation angles

BUT not every compment always present

Question 3

what are 4 prerequisites for self adaptation?

Answer 3

1. mu > 1 to carry different strategies
2. lambda > mu to create offspring surplus
3. comma selection to get rid of maladapted sigma
4. mixing strategy parameters by intermediary recombination on them

Question 4

using FSM to predict primes - what happened?

Answer 4

overfit like a bitch, always predicted ‘no prime’

>>> but apparently generally the concept works

Question 5

genetic programming: what is set during initialization?

> name to different types of init

Answer 5

genetic programming: maximum depth of trees is set at init

1. full method: every branch has depth = maxdepth
2. grow method: every branch has depth <= maxdepth

Question 6

genetic programming: whats bloat

> 2 countermeasures

Answer 6

bloat:

>survival of the fattest

> tree size increases over time

countermeasures

1. prohibiting variation operators that would deliver too big children
2. parsimony pressure

Question 7

differential evolution: how does differential mutation work?

Answer 7

differential mutation:

1. given a population of candidate solutions in R^N
2. for every candidate, create a new mutant vector
3. new mutant is produced by adding pertubation vector to the old one

Question 8

DE: how does the evolution cycle look like

Answer 8

DE:

1. population is list
2. create mutant vector pop
3. create trial vector pop
4. apply deterministic selection

Question 9

DE: how do variants differ?

Answer 9

DE:

> different variants have different base vectors

> described by DE/a/b/c

a: way to choose base vector (rand or best)
b: number of difference vectors to define pertubation vector
c: crossover scheme e.g. ‘bin’ for uniform xover

Question 10

particle swarm optimization: how does one member look like?

Answer 10

PSO:

one member is a 3 tuple of vectors

1. position vector
2. velocity vector
3. best positional vector of this particle in the past

Question 11

PSO: how does w affect the algorithms behavior?

Answer 11

PSO:

> wo controls exploration vs exploitation

if w > 1: velocities increase over time, swarm diverges

if 1 > w > 0: particles decelerate

Question 12

what is EDA, why is it usefeul?

Answer 12

EDA: estimation of distribution algorithms

> each algorithm has certain downsides, e.g. EA does not give any information about fitness landscape

> use model based EAs

>>> those try to model the distribution of the population and generate new solutions from them

Question 13

EDA in 6 steps

Answer 13

EDA

1. create init pop random
2. evaluate pop
3. select subpop (from best)
4. fit model to subpop
5. sample new pop from model
6. back to step 2