[backprojection]

That would be a fun project. If you can digitize the graph reliably, you can then take the discrete Fourier transform of the data, apply some denoising/thresholding to extract the main components. This would tell you if you're looking at sin(x), cos(x), cos^2(x), etc. If you have a training set, you could maybe start picking out more complex functions like abs(x), 1/x, etc.

[est]

I am referring to problems like this

http://math.stackexchange.com/questions/799790/simplest-know...

I don't think it's easy.

[geoffbrown2014]

I've been working on it in my spare time for 18 months and can do some rudimentary stuff. Graphing was the easy part. ;) I do a lot of material layout at work and need to maximize how many parts of a specific shape can fit on a single sheet of raw material. The hard part is coding the math. I'm using robust statistics/regression, DFT's, SVM's, PCA, nesting and clustering which are much harder to code robustly than anything else I've run into. The other hard part is selecting the correct algorithms for the correct problems. So far my intuition is pretty bad. The only saving grace is that the amount of available info about these problems has literally exploded in the last few years.