W3A2 - Trigger Word Detection

Hello everybody.

In the Trigger Word Detection assignment, I do not really understand the goal of using TimeDistributed layer for output layer of the model.
I also tested another model without using TimeDistributed with Dense layer (just use Dense layer for output layer) and the results of them are almost the same (loss and accuracy in training and validation set).

Can anyone explain that for me, please?..

Hey @Hoang_Trong_Vu,
Welcome to the community. I am assuming you have read the blog post mentioned in the assignment regarding the Time-Distributed Layer. If not, please check it out here.

In this blog, in the beginning itself, something regarding this has been mentioned:

Update Jun/2019: It seems that the Dense layer can now directly support 3D input, perhaps negating the need for the TimeDistributed layer in this example (thanks Nick).

When the assignment was created and/or updated, perhaps there was no such support offered by the Dense layer, and hence, the assignment makes use of the Time Distributed Layer.

However, note that Time Distributed Layer can still come in handy, specially for layers that don’t have the 3D support like Dense now has. For more information, you can check the official documentation here. I hope this helps.

Cheers,
Elemento

1 Like