To start with, I want to believe for a modification of an algorithm like evolutionary algorithm to be considered novel, it has to offer some benefit (like automation) and should not perform worse than the manually designed algorithm.
Base on the above, If I want to implement an evolutionary algorithm to design another evolutionary algorithm, I wouldn't know before hand whether the proposed method will perform as good or even better than the manual method or some other similar method.
I can only tell how the method performs after the completion of the implementation.
Is there a way to know before hand that a proposal will not bring some benefit, before putting energy and time into developing it?
Please, what do you advise in this situation from your experience?
Thank you in anticipation of your kind response.
If there are a lot of examples of how any modifications of the evolutionary algorithm improved or worsened its characteristics, then you can use AI to create a model that will recommend certain changes to achieve specific set parameters.
Usually the code can be optimised: same algorithm but less computing time, less memory (well, sometimes _far_ less). Now, for the quality of the results if you find a _sure_ way to improve it ((for a given search effort, of course, like a number of evaluations) publish it quickly!
Note that a different initialisation method might improve the algorithm on some problems, but not always. See for example
Omran, Mahamed G. H., Salah al-Sharhan, Ayed Salman, and Maurice Clerc. ‘Studying the Effect of Using Low-Discrepancy Sequences to Initialize Population-Based Optimization Algorithms’. Computational Optimization and Applications, 2013, 1–24.
Sometimes the problem can be reformulated. Using different formulae to take constraints into account or rewriting a discrete problem as binary.
Thank you very much for your responses. Let me give me a particular example. I developed a new crossover operator for genetic algorithm and only to realize at the end of the development it was performing worse than the traditional crossover operator.
Moreover, I devise some new way of configuring genetic algorithm (automatic configuration), after reading about similar approaches. However, after putting all the developmental effort; I got comments like does it perform better than the manual approach? How does it compare to other similar automated approaches (perhaps implying whether it performs better)?
- Badges
- Science topic
- Similar topics
- Correction