Zachary Litvinenko

https://pdfs.semanticscholar.org/9f5d/fa77f61a4dc87faa243f995e2092ddb3f521.pdf

Oh, sorry, it was a direct link to the article. For some reason they removed it.

can you try and feed the algorithm with less points

you want maxerror at something like 1.5-2 and less control points.

hmm I really don't like that wavy piece. did you use the code from the paper?

we can try some other curve fitting methods out there then. since you implemented beziers in milton, it's just a matter of 30 mins to add a new algo

I can just google for a bit and drop a couple potentially useful links here http://paperjs.org/tutorials/paths/smoothing-simplifying-flattening/

I have this issue with the coco dataset 

Maybe you need to have some time variable input into each of the oscillators so that you can synchronize and more precisely work with time. Which by default will be...

Also time plugged into the noise and everything that creates waves.