I have a complicated loss function `a(x, y)`

that takes a long time to compute. But I have 2 similar loss functions `b(x, y)`

and `c(x, y)`

that each capture some aspects of `a(x, y)`

.

How could I easiest find `f()`

and `g()`

such that `a(x, y) = f(b(x, y)) + g(c(x, y))`

?

I can sample any data `x`

, `y`

to try these functions. Would I do some curve fitting or evolutionary coding to get somewhere? I can try guessing but that will only get me so far.

Here are some quick values from inserting semi-arbitrary values to `a`

, `b`

, and `c`

: