The discussion in the video is restricted to inference, but how would one incorporate EP into the learning in a latent time series? Regrettably EP does not specify a variational lower bound of log likelihood (as compared to variational, EM-like methods), and, learning non-linear state transition functions in a time-series is non-trivial.
The discussion in the video is restricted to inference, but how would one incorporate EP into the learning in a latent time series? Regrettably EP does not specify a variational lower bound of log likelihood (as compared to variational, EM-like methods), and, learning non-linear state transition functions in a time-series is non-trivial.