Question

Qa: For fitness proportional, ranking and tournament selection separately, discuss
a) convergence speed, is it problem-specific or not?
b) can convergence speed be controlled and how?

Qb: What you possibly lose by increasing convergence speed?

Answer 1

Convergence Speed:

The speed at which a convergence sequence reaches it limit is called rate of convergence.

Qa . Fitness Proportionate:

a). In Fitness Proportionate Convergence speed is not problem specific because,Maintaining good fitness in the population is extremely crucial for the success of a Genetic Algorithms.

b). yes,convergence speed in fitness proportionate can be controlled.It can be justified as ,we have two implementations for fitness proportionate selection they are:

1. Roulette Wheel Selection:It consists of a fixed point.The region of the wheel which comes in front of the fixed point is chosen as the parent and the  process is repeated for remaining parts.

2. Stochastic Universal Sampling:It is quite similar to Roulette wheel method ,but instead of selecting one fixed point we are selecting multiple fixed points.all parents are selected in only one spin.  

Tournament Selection:

a). In Tournament selection , convergence speed is not a problem specific because, in K-Way tournament selection, we select K individuals from the population at random and select the best one from them to become a parent. The process is repeated until the selection the next parent.

b). yes,convergence speed can be controlled in tournament selection.It is done using repeated selection and the ability to reach the maximum fitness with lowest no. of generations is highest in deterministic

Ranking Selection:

a).In Ranking selection, convergence speed is also not a problem specific because we perform sorting to the population and then assign rank based Probabilities to them.

b).yes ,convergence speed can be controlled in ranking selection.It can be done maintaining steady pressure towards convergence.The selection of the parents depends on the rank of each individual and not depending on the  fitness. In this higher ranked individuals are considered more than the lower ranked individuals.

Qb . By Increasing the convergence speed , it is difficult to maintain the training iterations and causes to test loss.It may lead to diverge suddenly.