With traditional GA the order is crossover then mutation.

But there are some interesting research activity related to introducing of plasticity into evolved ANNs, which means that not only synaptic weights used to encode knowledge, but also some kind of plastic rule should be evolved as well, i.e. the ability for connections between neurons to change continually and automatically throughout life. Therefore, evolved plasticity rules incorporate lifelong experience even when underlying genome structure is already defined.

The majority of genetic algorithms were not devised to introduce plasticity into genome and mostly evolve fixed networks in which the final synaptic weights are encoded in the genome. But with introduction of additional genotypical instructions it's possible to evolve plasticity rules through direct genetic encoding.

My hypothesis is that this new genotypical instructions can be evolved in the existing population through mutations of already existing organisms, resulting in a more open-ended evolution that can evolve indefinitely more complex solutions. But it has to be tested.

For more details about concept of Evolved Plastic Artificial Neural Networks please refer to: https://arxiv.org/abs/1703.10371

dear researcher

there are many types of mutations,the selection of mutation and the applied of it depend on the problem you are working on and also depending on the rate of mutation .. you can see "The Role of Crossover on Optimization of a Function

Problem Using Genetic Algorithm" as an example. with best regards

Traditionally, it's crossover and then mutation. But it would be interesting to see if you mix up the order and had mutation followed by crossover. A simple implementation of GA with this new scheme should show if there are any differences from the original. Personally, I haven't come across a paper that has ever considered such a thing. Very interesting question.