/    /  Machine Learning- GENETIC ALGORITHM: MODELS OF EVOLUTION

GENETIC ALGORITHM: MODELS OF EVOLUTION

 

A genetic algorithm is a search type that was totally inspired by the theory of Charles Darwin, the theory of natural evolution-it covers the process of selection naturally where the individuals fit get selected for reproduction for the production of offspring to the next generation.

 Genetic algorithm covers the Darwinian model of evolution which states that natural selection can be possible as well as genetic mutation variation through reallocating and process of mutation.

However there are other models of evolution and lifetime adaption, they are as follows :

  1.       Lamarckian model and
  2.       Baldwinian model.

 

LAMARCKIAN MODEL :

The Lamarckian model defines that the traits of an individual get his or her lifetime can also be passed to its further offspring. It was then named after the French biologist – Jean Baptiste Lamarck. Though natural biology has excluded or disregarded the Lamarckism theory we know that only the information in a particular genotype can be transmitted.

But from a computation point of view, it has also been stated that the Lamarckian model also gives better results for some kinds of problems.

In the Lamarckian model, there is a local search operator which examines the neighbors getting new traits, and if a good chromosome is found then it itself becomes the offspring.

 

BALDWINIAN MODEL :

It was named after James Mark Baldwin in 189. It was an intermediate idea. As per the Baldwininan model, a chromosome can have the superficiality of encoding learning beneficial behavior types. Unlike the Lamarckian model, here it does not transmit the traits acquired to the next generation, and neither it completely ignores the traits acquired by the Darwinian model.

The Baldwinian model out of these extremes of two has the tendency of such kind of an individual for acquiring traits that are encoded except the traits of themselves.

In the Baldwinian model, there is a local search operator that exists that examines the neighbor on acquiring new traits, and if a good chromosome is found at the step, it then assigns only the improved  fitness factor to the particular chromosomes and it does not modify the chromosome itself. The ability on acquiring the trait though it does not get passed directly on to the fur=ture generations.

Therefore we often see a genetic algorithm with local search which is similar to the Memetic algorithm. Hence, one always must choose the correct and suitable models, either Lamarckian or Baldwinian to get decided what the individuals generate after a normal local search. Therefore, it generally follows the non-Darwinian theory or methodology of computing evolutionary procedures in machine learning for the guide of new individuals.

The networks which are neurally generated from Lamarckian theory states that they are small but require less time and to get designed. Thus, from every point of view, the Lamarckian operator gets improved by the fitness functions if used in early agreements. It was due to some part which gets passed to the next individuals which are fit and therefore the Lamarckian model is more suitable. 

 

Reference

GENETIC ALGORITHM: MODELS OF EVOLUTION