Lecture 3 - Luke Flashcards
Three properties of biological systems which make them key for inspiration
Robust
Complex
Adaptive
What is the use of evolutionary computation?
To copy the process of natural evolution to automatically find solutions for complex problems.
Define a population
Group of several individuals
Define diversity
Different characteristics between individuals
Define heredity
Characteristics passed down through generations
Define Natural selection
Survival of the fittest. Those with beneficial characteristics are more likely to survive and reproduce and thus make more offspring.
What is a genotype?
Genetic material that makes up the organism.
How is a genotype transmitted and what affects it?
Transmitted through reproduction and it is affected by mutations. It is not however directly affected by selection
What is a phenotype?
The manifestation of the organism, behaviour and appearance for example.
Genotype codes for..
Phenotype
What is DNA
Long molecule, double helix structure composed of the binding of four nucleotides.
What are the four nucleotides in DNA?
Adenine, cytosine, guanine and thymine.
What does the sequence of nucleotides code for?
Different forms of protein, mediated by mRNA.
What are the two regions of a gene?
Regulatory and Coding
Two types of cell replication, and a brief description.
Mitosis: Growth of cell
Meiosis: Production of sex cells
In what cells can mutations affect evolution?
SEX
Brief explanation of artificial evolution
Automatic generation of solutions to hard problems
Similarities between natural and artificial evolution
Genotype Phenotype (computer program, object shape, electronic circuit, robot, etc.) Population Diversity Selection Inheritance
Differences between natural and artificial evolution
With artificial there is an expected improvement between initial and final solution.
Selection in artificial is determined by a predefined measure, whereas in nature it is primarily survival and reproduction.
Approach to an evolutionary algorithm
Make a genetic representation Build a population Decide on a specific fitness function Choose selection method Choose crossover and mutation Choose a data analysis method
Three conditions for stopping evolutionary algorithm
Max fitness value found
solution found is good enough
no fitness improvement (stagnation)
Operators and Operands involved in tree-based representations?
Operators - function set (multiplication, if-then, log etc)
Operands - terminal set (constants, variables, sensor readings etc.)
What is sufficiency with regards to tree based representations
The elements in the function and terminal sets must be sufficient to generate a program that solves the problem
What is closure with regards to tree based representations
All functions must accept all terminals in Terminal set and outputs of all functions in Function set
How do you decide on initial populations?
Large enough to cover the problem space, but not too large in terms of evaluation costs
Describe the initial populations for trees
Trees are built recursively starting from root: root is randomly chosen from function set; for every branch, randomly choose among all elements of function set and of terminal set; if terminal is chosen, it becomes leaf; set maximum depth of tree.
What is the role of the fitness function
Evaluates the performance of a phenotype with a numerical score
(You get what you evaluate!)
What is subjective fitness?
It is used to select the phenotype by visual inspection. Used when properties are not quantifiable, such as the aesthetic properties.
What is the difference between explicit and implicit fitness functions?
Explicit fitness functions evaluate every detail and implicit is more general
What is the purpose of the selection element of artificial evolution?
To make sure that better individuals make comparatively more offspring
Selection pressure is inversely proportional to what?
Number of individuals selected
What does a high selection pressure suggest in terms of diversity?
A loss of diversity, and premature convergence
What are the problems of proportional selection?
If there are uniform fitness values, this results in a random search.
If a few individuals have high fitness values then this leads to a high selection pressure
Five types of selection
Proportional rank based tournament truncated rank based replacement selection
How does truncated rank based work
Only the best x individuals are allowed to make offspring and each of them makes the same number of offspring: N/x, where N is the population size.
How does the size of the tournament affect selection pressure?
Larger K higher selection pressure
Three types of replacement selection
Generation replacement (old population entirely replaced by offspring) Elitism (maintain n best individuals from previous generation to prevent loss of best individuals by effects of mutations or sub-optimal fitness evaluations) Generation rollover (insert offspring at the place of worst individuals)
Three types of crossover
Single point
Uniform
arithmetic
What is a fitness landscape
PLot of fitness values associated to all genotypes
How to estimate the ruggedness of real landscape?
Sample random genotypes: if flat, use large populations
Explore surroundings of individual by applying genetic operators in
sequence for fixed number of times: the larger the fitness improvement the easier is to evolve
What can the stagnation of fitness function suggest?
Best solution found or premature convergence
Applications of evolutionary algorithms
Evolution of Neural Networks Evolution of Cellular Automata Evolution of hardware (tree based encoding) Evolution of Controllers Evolution of laws of physics
