The method of optimization probably has little to do with the interpretability of the model. Gradient optimization (i.e. backpropagation) can be used to fit log

The method of optimization probably has little to do with the interpretability of the model. Gradient optimization (i.e. backpropagation) can be used to fit logistic regression models, which are interpretable. Evolutionary algorithms can be used to fit deep neural networks, and will not make it more interpretable. I'd say the least intepretable systems are biological, which are the outcome of an evolutionary process