There isn’t much to explain about these topics since they are in some sense a spin-off of GD created to cope with large amounts of data while achieving reasonable performance.
I have provided some links to resources that should help but writing full articles on such trivial topics would be —…
Another great question that really made me think.
I think that, NEATs are great but from what I could infer from the video and after reading a few lines of the paper is that:
I’m sorry but I’m unable to understand what you mean by, “optimize the layour of the ANN.”
From, what I can infer, you’re talking about optimizing just a layer or a few(in a many layer NN). This is definitely possible but you would be leaving out the other layers which also have a major contribution to the NN.
On second though, there might be a way of avoiding the problem you mentioned in the second question.
We can do the following:
Run our GA to optimize the hyper-parameters. After obtaining the hyper-parameters found by the GA(let’s name these HP1) we’ll run the GA again. Since, we’re initializing…
First of all, using a GA we can at least be sure that our hyper-parameters are close to the optimal. It is almost impossible to find the optimal hyper-parameters for any model. If we could, then every model would attain 100% accuracy. If that were the case, I wouldn’t have written an article about it.