Jag tror min idé är relaterad till P vs. NP problemet. Jag antar att ?alla problem kan skrivas som ett maskin inlärnings nätverk. Då min idé ger en sådan aning att det finns en modell av problemet som löses med ?bestämt antal beräkningar.
Taget från min blogg.
http://peroglyfer.se/2017/11/22/machine-learning-idea-complexity-recognition-from-smoothed-weight-matrices-is-the-problem-easy-or-ha rd/
Machine Learning Idea - Could You Solve The Complexity Problem From Recognition Of In Process Smoothed Weight Matrices?
If anything would solve the complexity problem it would be a network.
I mean.
In the brain we think a little bit about a problem then decide if its difficult or not. So inspired by this. Could a machine learning network method solve ?any such problem of complexity questions?
So my idea is.
If a problem of high complexity requires a complex weight matrix. Then I think you can begin with iterating and judge it by using Gaussian filter in the process of creating the weight matrices.
First I guessed that by looking at the iterated softened weight matrix or using them in a classification network I could use another network to determine a complexity value or vector/image.
My second guess is that. If I get a smooth somewhat converged weight matrix. Then train on an additional set of samples from the ?test set and then compare the change of the weight matrix to the original. I guess this would reflect how difficult the problem is. The ?more change. The more problematic the problem.
Could this be a way to predict how hard a problem is? If so then with the complexity images I think an algorithm can suggest a corresponding network setup model.
