Project approach consultation - RF / DL / …?

Hi :slight_smile: hopefully this isn’t too long. I tried accentuating things in bold.

I have a specific project in mind I wish to do using AI, but I’m unsure which method to use, and where to focus my time in terms of the different courses and lessons available (ML / DL / etc.). The details are less important, but I’d be happy to explain more if needed. For now I’ll just describe what I’m trying to achieve in general.

I have sampled points on a general mult-dimensional tensor, and the goal is to find a mapping function of these points that achieves good results in terms of a criteria I set it (MSE), under a certain constraint - some sort of upper bound limitation on the mapping. The details are less important, the bottom line is that I’m trying to find a mapping for these points g(•)=? that minimizes MSE under a constraint. So, for X sampled points, I need to find X points that correspond with their mapping. Everything is continuous of course - the sampled points and their mapping.

I’ve already built a program that does this using scipy.optimize.minimize, which uses an iterative method for constrained nonlinear optimization - uses gradient decent to find a local minimum that achieves the criteria under the constraint.

I want to try to use and test the performance of a neural network, or random forest, or something along those lines using AI, but I got a bit lost in the videos I started watching, and I did not find a suitable example that was close to what I’m trying to achieve. In short, I’m unsure which algorithm I should be using, and if AI is suited for this task at all, or is what I’ve already built the “standard” way of approaching this task, and that’s that.

I’d like to ask which approach you think would suit my problem (RF? DL? etc.), and if you believe that these approaches could possibly outperform the straight-forward approach I already implemented using the iterative method, which suffers from the unwanted local minimum phenomenon, and is relatively slow in general (especially in high dimensions).