How to do piecewise optimization if given model but not the fusage points?

37 Views Asked by At

Fitting data with piecewise models given fixed points of fusage, or intervals of validity with given number of continous moments across the models is a well explored problem with hundreds if not thousand of methods. But what if we don't know the points of fusage of the models? Can we find a method to adaptively find the points of fusage or intervals of validity for each model?

Here is example of a noisy function which we want to fit say two polynomials $p_1(x),p_2(x)$ to. How can we choose the fuseagepoint $x_0$ to get as good a fit as possible given we enforce $$p_1^{(k)}(x_0) = p_2^{(k)}(x_0), \forall k \in [0,n]$$

or a softer variant, penalty terms added to the piecewise fitter (assuming an optimization):

$$\sum_k\epsilon_k\|p_1^{(k)}(x_0) - p_2^{(k)}(x_0)\|_2^2$$

enter image description here