context_recommendation icon indicating copy to clipboard operation
context_recommendation copied to clipboard

Published 20 hours ago • verified publisher icon rockingdingo

About multi-head

Open EagleYing opened this issue 4 years ago • 0 comments

The article describes that the k hidden units in the middle of the multi-head AutoEncoder can obtain local context information, but why each hidden unit uses all the input mask(x), which is equivalent to calculating the traditional DAE K times. Does this make any sense?

EagleYing avatar Dec 08 '20 13:12 EagleYing