-
Notifications
You must be signed in to change notification settings - Fork 9
Open
Description
Hi, I have a doubt regarding this paragraph:
This is now a little weird too, because the algorithm starts and ends with independent updates to the parameters, so going from time t to time t+1 involves two successive parameter updates--why not combine them into one parameter update? We can do this by subtracting mu*m_tm1p1 at the definition of x_t, where m_tm1p1 is m_tm1 at the next timestep t+1, which is obviously m_t at the current timestep:
Can you please elaborate on this? I couldn't understand this step.
Thanks!
Reactions are currently unavailable
Metadata
Metadata
Assignees
Labels
No labels