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
I've successfully trained an STN with an embedded mapping of the "Drive" parameter, that is ~2.5x more efficient than the NR8 mode. It seems like training with embeddings of the "Saturation" and "Bias" parameters, will be difficult, and even if it can be done, he resulting network will most likely be computationally too expensive.
Instead I'm planning to train STNs for 10 unique "Bias" parameter values and 20 "Saturation" parameter values, and choose the correct network based on the real-time parameter choices.
Implement a hysteresis mode using a state transition network (see, Parker, et. al., 2019).
The text was updated successfully, but these errors were encountered: