This looks like a coding language that can take distributions, and output co-ordinates or arguments, and then perform algorithmic transforms on them. i think. Way over my haircut. Looks like an efficient way to work with sets tho.
“These source-to-source transformations perform exact inference as well as generate probabilistic programs that compute expectations, densities, and MCMC samples. The resulting inference procedures run in time comparable to that of handwritten procedures. ”
This looks like a coding language that can take distributions, and output co-ordinates or arguments, and then perform algorithmic transforms on them. i think. Way over my haircut. Looks like an efficient way to work with sets tho.
“These source-to-source transformations perform exact inference as well as generate probabilistic programs that compute expectations, densities, and MCMC samples. The resulting inference procedures run in time comparable to that of handwritten procedures. ”
https://arxiv.org/abs/1603.01882
If they actually achieved what they claim—a calculus for directly obtaining inference algorithms from input models—then it is very cool.
My gut mathematical feeling is that the full solution to that problem is intractable, but even a partial semi-solution could still be cool.