I’m trying to figure out a super battery. The philosophical abstract part. In that I’m using machine learning.
Maybe you need better machine learning for a super battery. So here goes.
The idea is to have “error correction” on each layer. So what error inputs do we have if we only got a target in the last layer?
What I guess is that you can you compare each layer with a ?image denoise filter. This way you get a layer that approaches denoise(layer). I will try assigning L1 = Denoise(L1) and so on.
With this I think I took some convergence problems of some many layer models.