Download as docx, pdf, or txt
Download as docx, pdf, or txt
You are on page 1of 2

CS1AC16 OTHER MODERN METHODS

Evolutionary computing – a different method of problem solving – inspired by biological processes and
Darwin's theory of evolution
Basic idea:
o Start with a problem and a number of possible solutions
o Each of the solutions is a member of the population
o Assess how good each solution is
o Use an algorithm to breed some new solutions which are then incorporated into the
population
This is based on the Darwinian concept that better members of the population survive and breed and
their children are even better.
Issues:
o How to represent a population? - the simplest way is a series of bits but it could also be
numbers or some other way
o How to calculate fitness? - this is problem dependent
o How to breed? - suitable genetic operators are needed
o How to incorporate the next generation? - one way of doing this is by replacing the
weakest members with new ones

Weighted roulette wheel selection – a way of choosing the population members to breed. There is
randomness involved but good solutions are more likely to be chosen.
You form a wheel with f slots (equal to the total fitness of the population) and each member is allocated
the number of slots according to their fitness – better solutions have more slots so they're more likely
to be chosen

Breeding – can be done with one member or 2


 Inversion – pick 2 random bits in a solution and swap them
 Mutation – invert 1 random bits (should not be used too often)
 Crossover – combine parts of 2 solutions
To form the next generation you remove the weakest members of the population and replace them with
the new ones.
Over time, the best solution is found.
Problems:
o Genetic algorithm take a lot of time
o Can lead to many similar but not the best solutions

Agents
Each member of a genetic algorithm population is called an agent. Agents are potential solutions to a
problem. They can also be parts of a solution – to form a full solution, many of them have to be
connected (like in neural networks). They are simple entities that interact with each other and with the
environment. They can be software and hardware and are also inspired by nature.

Swarm intelligence
This is a form of artificial intelligence that emerges when many agents work together, forming
something useful as a result.
Each agent obeys some simple rules and there is no centralized intelligence.
Example: ant colony optimization – inspired by how a colony of ants searches for resources:
o Initially the ants walk around randomly
o If they meet another ant that has found resources, they join it
o Eventually the ants follow a trail to the resources

Stochastic diffusion search


 There is a series of agents and each agent is a potential solution to part of a problem. The
agents communicate about whether they have a possible solution.
 At each iteration an agent tests a different aspect of its possible solution
 When an agent with no solution meets an agent with one, it takes the possible solution,
otherwise it chooses randomly
 Over time the agents cluster around possible solutions

Hybrid systems
Some methods work better than others for certain problems. Sometimes we have systems comprising
different methods – each working on where it works best.

You might also like