11 ms·
they do not use priors or expert belief. they solve the system id problem directly - which in most problems is unresolvable
by starwars11 7y ago
they do not use priors or expert belief. they solve the system id problem directly - which in most problems is unresolvable
- bananaquant 7y agoI'm just working through the math in the paper right now. To me, it looks like all of the formulas could be expressed in a probabilistic PL and MCMC could be used to estimate the probabilities of both latent state and parameters. Would there be some particular roadblocks to this approach?
- yalph 7y agoWhat does solving system id mean?
- starwars11 7y agoestimating model parameters. normally people use heuristics because it’s an unreasonable problem for linear dynamical systems without additional assumptions. here, they’re able to produce analytical results and finite sample analysis
- starwars11 7y ago*unresolvable