You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
For state space models, we typically don't have reverse kernel because the state dimension grows over time. This example will greatly illustrate how to deal with growing-dimensional variables in JAX. The trick will be to prepare a full dimensional variable and perform index update in each smc step.
The text was updated successfully, but these errors were encountered:
Neural Adaptive SMC, Gu etc. is a nice framework that allows us to train proposals for non-linear state space models. We can use forward KL in a nested variational inference scheme because both derivations provide similar grad estimations.
For state space models, we typically don't have reverse kernel because the state dimension grows over time. This example will greatly illustrate how to deal with growing-dimensional variables in JAX. The trick will be to prepare a full dimensional variable and perform index update in each smc step.
The text was updated successfully, but these errors were encountered: