2. Branch of AI
Geared towards taking input and yielding
patterns or predictions
Use data to capture uncertainty in underlying
probability distributions
3. 1959 – Arthur Samuel: "Field of study that
gives computers the ability to learn without
being explicitly programmed".
Machine learning != data mining
5. Machine learning technique
Used to optimize a program’s ability to
perform some kind of task
Inspired by biological evolution (i.e.
Darwinian evolution)
6. Explicit + implicit parallelism
Gracefulness
Knowledge about evolutionary processes
(brain processes are harder)
7. Natural evolution is too slow
Silly argument – generations are much quicker in
computers than biology and so many generations
have given great process
Waddington (1967) says GP algorithms are
too simple to get to the same type of
complexity as real evolution
Answer: GP works for efficiency in a wide range of
problems
10. Let us try to find an algorithm, where the
operators are + and – to get to a target
number, which we call x. We will restrict
ourselves to 5 “spots” and single digit
numbers
E.g. Let x=10
Solutions would be:
▪ 0+2+8, 1+9-0 , 8+3-1, …
11. Create a population of individuals
1+3-4
++23-
0+4+1
983-2
…
12. How far off of our target are we?
1+3-4 (fitness=10-0=0)
++23- (fitness=-100 since invalid)
0+4+1 (fitness = 10-5=5)
983-2 = 3-2 (fitness = 1)
…
13. Select the most fit individuals to breed (or
some other algorithm)
1+3-4 (fitness=0)
++23- (fitness=-100 since invalid)
0+4+1 (fitness = 5)
983-2 = 3-2 = 0+3-2 (fitness = 1)
…
14. Crossover (take part from each)
0+4+1
0+3-2
=> 0+3+1 = 4 (fitness = 4)
Mutate with some probability (change one
number of operator
Put new individual into population
Remove least fit individual (the invalid one)
16. Keep going until you have a perfectly fit
individual
Sometimes the trait you need isn’t in the
population
That’s why mutation exists
Applications are much wider than this, and
can generate real programming language
code
17.
18. http://en.wikipedia.org/wiki/Machine_learnin
g
http://en.wikipedia.org/wiki/Genetic_progra
mming
D. E. Goldberg and J. H. Holland, “Genetic
Algorithms and Machine Learning,” Machine
Learning, vol. 3, no. 2, pp. 95–99, 1988.