Introduction
Genetic Algorithm is based on the simulation of natural environment. In the mathematical model, it is a kind of method to get the result through random series. So in other hands, if we design the algorithm through a concrete way and then translate the algorithm in an abstract way, the efficiency of GAs can be increased!
Main Objection
1. Designing a novel style of GA combined with bio-inspired methods and statistic theory.
2. Try to find a method to transfer bio-inspired GAs into statistic style.
3. Parallel style of this algorithm.
4. Try to find the hidden connected between PF, PSO & GA.
Over View:
1.Designing a novel style of GA combined with bio-inspired methods and statistic theory
GA is a kind of algorithm try to optimize target function through complex simulation. Although the ability of individual is very limited, but the whole system is complex and has a strong ability to deal with optimization. The methods for constructing the uncertain environment is bio-inspired. However, it may cost very large source because the process of simulation. The first goal of this project is trying to design GA from a statistic way. Although it may be less complex than bio-inspired GA, but the performance may still follow our demands.
2. Try to find a method to transfer bio-inspired GAs into statistic style.
There are many exist GA in the world. How to transfer those existing methods is a problem can give us uncountable benefits. So the next objection is trying to find a way to transfer bio-inspired GAs into statistic style to increase their efficiency. And this also can benefit the widespread of these kinds of the algorithm. The simplest example is given in the Current Result. The most difficult problem in this project is the complex of density function created by GAs. Currently, I try to find a way to weak the density function so that the universal methods can be found even the GA's density function is very complex.
3. Parallel style of this algorithm.
Comming Soon...
4. Try to find the hidden connected between PF, PSO & GA.
Comming Soon...
Related Work
[1] has put forward an algorithm called abstract evolutionary algorithm. And also some researchers try to describe the basic theory about it. It is very similar like this algorithm. But their work mainly based on finding the abstract type algorithm. My work mainly focuses on transfer existing method into an abstract way and adaptive controls it. In other hands, my work also tries to find a hint from PF theory [2] & MCMC theory [3], which [1] doesn't focus on.
Novel Points
1. Using abstract GAs can save the cost. So it is good at dealing with very large amount of data
2. The abstract type of GA can easier to adaptive control than normal GAs.
3. The lines of abstract GA is less than normal GA.
4. The normal GAs can be translated into abstract style. So a universal algorithm can be designed based on exist method.
Current Result
The abstract Population style algorithm:
In this style of GA, the population is density function (non-parameter). In follow figure, it shows an algorithm described in my publication. The population still noted by Mixture Gaussian Density Function.
Compared with normal style genetic algorithm
In [4], compared with normal algorithm, actually there are very little different between two algorithms' result. But the abstract type genetic algorithm may be very efficiency when meet very large amount of data. Because the complexity of abstract style GAs may less than normal style GAs. The performance of both style algorithms in real project can be seen in follow figure.
The Challenge
The combination of genetic operation
The most power property that abstract genetic contained is the operations can be combined (For example, we can combine the crossover operation & mutation operation into one operation). But it is very difficult to create an algorithm that can be widely used in real problems because the random series that created by GAs always very complex.
Download the Reference
Return to Research Page
|
