MastersWork.PresentationOutline History
Hide minor edits - Show changes to output
Changed line 64 from:
* What works well for one GA may completly ruin another one
to:
* What works well for one GA may completely ruin another one
Changed line 9 from:
* A popluation of solutions are tested to see which are the best, and from these new solutions are created and tested.
to:
* A population of solutions are tested to see which are the best, and from these new solutions are created and tested.
Changed line 34 from:
* '''''[A: is there a good example to use here?]'''''
to:
* '''''[A: is there a good example to use here? R: Have yet to come up with one. Suggestion welcome :) ]'''''
Added line 48:
'''''[Can use right hand side of diagram to explain the evolution path]'''''
Deleted lines 14-15:
Changed lines 22-23 from:
to:
http://www.kallisti.net.nz/~robin/genphen.png
[[<<]]
[[<<]]
Changed lines 47-49 from:
'''''[Insert example diagram of Gray code evolution]'''''
to:
http://www.kallisti.net.nz/~robin/gray.png
Changed lines 16-18 from:
http://www.kallisti.net.nz/~robin/crossover.png http://www.kallisti.net.nz/~robin/mutation.png[[<<]]
to:
http://www.kallisti.net.nz/~robin/crossover.png http://www.kallisti.net.nz/~robin/mutation.png
[[<<]]
Changed lines 16-17 from:
http://www.kallisti.net.nz/~robin/crossover.png http://www.kallisti.net.nz/~robin/mutation.png
to:
http://www.kallisti.net.nz/~robin/crossover.png http://www.kallisti.net.nz/~robin/mutation.png[[<<]]
Added line 16:
http://www.kallisti.net.nz/~robin/crossover.png http://www.kallisti.net.nz/~robin/mutation.png
Changed lines 25-26 from:
* In biologial systems: genotype = genes, phenotype = body, fitness function = environment
* The single most important part of a GA, as it defines its purpose. [A: more important than representation?]
* The single most
to:
* The single most important part of a GA, as it defines its purpose. '''''[A: more important than representation? R: I think so, without a good fitness func the GA isn't going so produce any relevant results, without a good representation you are going to impair the results you do get.]'''''
* In biologial systems: genotype = genes, phenotype = body, fitness function = environment.
* In biologial systems: genotype = genes, phenotype = body, fitness function = environment.
Changed line 32 from:
* [A: is there a good example to use here?]
to:
* '''''[A: is there a good example to use here?]'''''
Changed line 5 from:
* How representation affects the evolvability of Genetic Algorithms [A: odd capitalisation]
to:
* How representation affects the evolvability of genetic algorithms.
Changed line 8 from:
* They [A:roughly] mimic Darwinian evolution in order to quickly find a good solution to a problem.
to:
* They roughly mimic Darwinian evolution in order to quickly find a good solution to a problem.
Changed line 12 from:
* Ideas take [A: taken] from natural evolution.
to:
* Ideas taken from natural evolution.
Changed line 14 from:
* [A: what about mutation - its mentioned below]
to:
* Mutation: making a small random change to the genotype.
Changed lines 25-26 from:
* It is this that determines what solution is best [A: redundant given above point?] .
* [A: genotype = genes, phenotype = body, fit func = environment]
* [A: genotype = genes, phenotype = body, fit func = environment]
to:
* In biologial systems: genotype = genes, phenotype = body, fitness function = environment
Added line 14:
* [A: what about mutation - its mentioned below]
Changed lines 25-26 from:
* It is this that determines what solution is best.
* The single most important part of a GA, as it defines its purpose.
* The single most important part of a GA, as it defines its
to:
* It is this that determines what solution is best [A: redundant given above point?] .
* [A: genotype = genes, phenotype = body, fit func = environment]
* The single most important part of a GA, as it defines its purpose. [A: more important than representation?]
* [A: genotype = genes, phenotype = body, fit func = environment]
* The single most important part of a GA, as it defines its purpose. [A: more important than representation?]
Added line 33:
* [A: is there a good example to use here?]
Changed line 42 from:
* An alternative to standard binary coding
to:
* An alternative to standard binary coding.
Changed line 8 from:
* They mimic Darwinian evolution in order to quickly find a good solution to a problem.
to:
* They [A:roughly] mimic Darwinian evolution in order to quickly find a good solution to a problem.
Changed line 12 from:
* Ideas take from natural evolution.
to:
* Ideas take [A: taken] from natural evolution.
Changed line 5 from:
* How representation affects the evolvability of Genetic Algorithms
to:
* How representation affects the evolvability of Genetic Algorithms [A: odd capitalisation]
Changed line 17 from:
* One of the most important parts of GAs.
to:
* One of the most important parts of a GA.
Added lines 22-26:
!!!Introduction to GAs: Fitness
* The fitness function takes the phenotype of a potential solution and returns a number saying how good it is.
* It is this that determines what solution is best.
* The single most important part of a GA, as it defines its purpose.
* The fitness function takes the phenotype of a potential solution and returns a number saying how good it is.
* It is this that determines what solution is best.
* The single most important part of a GA, as it defines its purpose.
Added lines 64-66:
* Not a lot of papers discuss trying different representations.
* Many don't even say what the representation is (making their results unreproducible).
* In general, representation is something that is worked out, ignored, and most effort goes in to tuning mutation and crossover rates.
* Many don't even say what the representation is (making their results unreproducible).
* In general, representation is something that is worked out, ignored, and most effort goes in to tuning mutation and crossover rates.
Changed lines 68-70 from:
!!!Future: Extraction of Features
!!!Cage Fight: Testing the Features
!!!Cage Fight: Testing the Features
to:
!!!End Result
* Ideally, a set of "If your problem can be reduced to this canonical problem, then you should do it this way".
* Ideally, a set of "If your problem can be reduced to this canonical problem, then you should do it this way".
Changed line 61 from:
* If we work with these general problems, then an improvement may help for all similar case.
to:
* If we work with these general problems, then an improvement may help for similar cases.
Deleted line 64:
Added line 39:
Added lines 49-52:
* What happens if the search space is larger than the solution space?
* That is, more than one genotype maps to one phenotype.
* Obviously there may be more searching to do, however...
* ...there may also be a larger number of easy paths to follow to the best solution.
* That is, more than one genotype maps to one phenotype.
* Obviously there may be more searching to do, however...
* ...there may also be a larger number of easy paths to follow to the best solution.
Added lines 55-61:
* This isn't very cut-and-dried.
* What works well for one GA may completly ruin another one
* However, there are a lot of common situations:
** Searching for a set of parameters that optimise a numeric function
** Optimising a sequence of operations (Travelling Salesman Problem)
** Building an algorithm for finding a general solution to some other problem
* If we work with these general problems, then an improvement may help for all similar case.
* What works well for one GA may completly ruin another one
* However, there are a lot of common situations:
** Searching for a set of parameters that optimise a numeric function
** Optimising a sequence of operations (Travelling Salesman Problem)
** Building an algorithm for finding a general solution to some other problem
* If we work with these general problems, then an improvement may help for all similar case.
Added line 65:
Changed line 5 from:
to:
* How representation affects the evolvability of Genetic Algorithms
Added lines 34-39:
* An alternative to standard binary coding
* Has 'adjacency', that is, there is an unbroken path from any point to the optimal point that is always increasing.
!!!Gray Code
'''''[Insert Table of Gray code to binary mapping]'''''
'''''[Insert example diagram of Gray code evolution]'''''
* Has 'adjacency', that is, there is an unbroken path from any point to the optimal point that is always increasing.
!!!Gray Code
'''''[Insert Table of Gray code to binary mapping]'''''
'''''[Insert example diagram of Gray code evolution]'''''
Added lines 42-45:
* How is the result of a particular value on one place on the genotype affected by another place?
* What happens when it gets split up by crossover?
* Orthoganality should be maximised, but it is not always possible to have it perfect.
* How much influence does it have?
* What happens when it gets split up by crossover?
* Orthoganality should be maximised, but it is not always possible to have it perfect.
* How much influence does it have?
Changed line 31 from:
* A set of instructions (assembly-like, execution tree) '''''what is the proper name for a tree structure of instructions? - I don't think there is one formal name - the genetic programming book by Kobsa (?) uses them all the time, it might be good to check there and see if there is any term he defines. A'''''
to:
* A set of instructions (assembly-like, execution tree) '''''what is the proper name for a tree structure of instructions? - I don't think there is one formal name - the genetic programming book by J R Koza uses them all the time, it might be good to check there and see if there is any term he defines. A'''''
Changed line 31 from:
* A set of instructions (assembly-like, execution tree) '''''what is the proper name for a tree structure of instructions?'''''
to:
* A set of instructions (assembly-like, execution tree) '''''what is the proper name for a tree structure of instructions? - I don't think there is one formal name - the genetic programming book by Kobsa (?) uses them all the time, it might be good to check there and see if there is any term he defines. A'''''
Changed lines 1-48 from:
to:
!!Presentation
This is the outline and text of my presentation which will be fleshed out with more information as it is worked out.
!!!Title Page
!!!Introduction
* Genetic algorithms (GAs) are a powerful solution to optimisation problems.
* They mimic Darwinian evolution in order to quickly find a good solution to a problem.
* A popluation of solutions are tested to see which are the best, and from these new solutions are created and tested.
!!!Introduction to GAs: Crossover and Mutation
* Ideas take from natural evolution.
* Crossover: combining two good solutions into a hopefully better one.
'''''[Include simple diagram]'''''
!!!Introduction to GAs: Genotype->Phenotype
* One of the most important parts of GAs.
* Genotype: the data that is being operated on by the GA with crossover and mutation.
* Phenotype: the expression of the genotype in a form to be tested for fitness.
'''''[include simple diagram]'''''
!!!Importance of the Genotype to Phenotype mapping
: :''Representation scheme decides not only the resolution of the obtained solutions and the size of the search space, but also the principle of translating and understanding the primitive problem.''
* A good representation leads to a quicker time to an ideal solution, and maybe a better solution.
* A bad representation may make it difficult, if not impossible, to reach a good solution.
!!!Different representation types
A few different ones:
* Search for a set of values (plain numbers).
* A sequence (ordered numbers, with or without repeats).
* A set of instructions (assembly-like, execution tree) '''''what is the proper name for a tree structure of instructions?'''''
!!!Simple optimisation: Gray code
!!!Larger Scale Optimisations: Orthoganality
!!!Larger Scale Optimisations: Redundancy
!!!Problems Working It Out
!!!Literature Breakdown
!!!Future: Extraction of Features
!!!Cage Fight: Testing the Features
----
<|HomePage|>
This is the outline and text of my presentation which will be fleshed out with more information as it is worked out.
!!!Title Page
!!!Introduction
* Genetic algorithms (GAs) are a powerful solution to optimisation problems.
* They mimic Darwinian evolution in order to quickly find a good solution to a problem.
* A popluation of solutions are tested to see which are the best, and from these new solutions are created and tested.
!!!Introduction to GAs: Crossover and Mutation
* Ideas take from natural evolution.
* Crossover: combining two good solutions into a hopefully better one.
'''''[Include simple diagram]'''''
!!!Introduction to GAs: Genotype->Phenotype
* One of the most important parts of GAs.
* Genotype: the data that is being operated on by the GA with crossover and mutation.
* Phenotype: the expression of the genotype in a form to be tested for fitness.
'''''[include simple diagram]'''''
!!!Importance of the Genotype to Phenotype mapping
: :''Representation scheme decides not only the resolution of the obtained solutions and the size of the search space, but also the principle of translating and understanding the primitive problem.''
* A good representation leads to a quicker time to an ideal solution, and maybe a better solution.
* A bad representation may make it difficult, if not impossible, to reach a good solution.
!!!Different representation types
A few different ones:
* Search for a set of values (plain numbers).
* A sequence (ordered numbers, with or without repeats).
* A set of instructions (assembly-like, execution tree) '''''what is the proper name for a tree structure of instructions?'''''
!!!Simple optimisation: Gray code
!!!Larger Scale Optimisations: Orthoganality
!!!Larger Scale Optimisations: Redundancy
!!!Problems Working It Out
!!!Literature Breakdown
!!!Future: Extraction of Features
!!!Cage Fight: Testing the Features
----
<|HomePage|>
Page last modified on January 12, 2005, at 02:50 AM