Parameter control (Self Adaptive) Optimization with Hill Climbing
Mohit Kumar Goyal (1583836)
Reason of selecting topic:
There are many scientific and engineering problems related to optimization which requires specific optimum input parameters in order to optimize the system response, like its transfer function and derivatives etcetera is not known and the measures might be incomplete and distorted by noise. This makes such problems difficult to be solved by traditional methods. Here, some optimization algorithms like Genetic Algorithm or Simulated Annealing can offer a solution, but:
1. Because of the lack of a standard methodology for matching a problem with a suitable algorithm and for setting the control parameters for the algorithm; practitioners often find difficulties to implement optimization techniques.
2. Parameters tuning leads to additional computational costs because of time consuming trials and finding suitable control parameter settings, will require to carry out a large number of experiments or error tests.
3. The other major concern for this is that a practitioner, who wants to apply an algorithm to a specific problem, and who has no experience with optimization algorithms, would need to become an expert in optimization algorithms before being able to choose a suitable algorithm for the problem at hand.
These above points becomes complex for a scientist or engineer, who simply wants to use optimization algorithms as a tool.
To overcome the above mentioned problems, I would like to present a proposal on optimization techniques which do not require any manual tuning of these control parameters. Instead choose an optimization algorithm that would control its parameters by itself based on the quality of a fitness function and it can try effectively a number of combinations to optimize the given problem.
There are two main approaches to the elimination of parameters in optimisation algorithms:
a) Parameter Control
b) Parameter Tuning
(a) Parameter tuning involves finding good values for the parameters before the algorithm is run and then using these values during the run.
(b) In Parameter Control, one starts with certain initial parameter values. These initial values are then adjusted, during run-time, in a number of ways. The manner in which the values of the parameters are adapted at run-time is the basis of Eiben’s classification of Parameter Control into three different sub-categories. These sub- categories are:
b) Adaptive and finally
My aim is to explore and discuss more about recent parameterless techniques and finally focusing on Self Adaptive parameter control so as to submit my review on papers related to parameter control self adaptive techniques on Hill-Climbing. I will exhibit how parameter control self adaptive optimisation helps to control the step size by itself which is the crucial to the success of the Hill Climbing algorithm. Furthermore, I will also discuss the various pseudo algorithms which helps to optimise the result without any manual interference. In the end I will conclude that how these techniques can save cost as well time as compare to previous ones.