Pointer Networks

[@cesar73silva]

Very good video, thanks! Drinking game for the viewers: take a shot every time he says "basically"

[@9739865097]

How is the variable d_i in the pointer network computed. For example, u^1_2 requires d_2. How is d_2 computed?

[@dagmawimoges6080]

amazing video. It is very helpful on its own, however, a rough overview of the code would be a plus.

[@KrishnaDN]

Sure Dagmawi, from next paper onward I will try to explain the code also. Thanks for the input

[@sareek007]

in results section, u said n represents tour length, but I guess n represents number of of nodes(cities) and numbers in other columns represent tour length by A1, A2, A3 paper and Ptr-Net model. Am I right on this??

[@BalaguruGupta]

the tour length will be n-1 and the number of nodes will be n. The n value is length_of(A1, A2, A3) = 3. The tour length will be n-1 => 3-1 = 2

[@mariamgarba1416]

Clearly explained,thank you. If you don’t mind, could you share a working code of this paper if it’s available?

[@KrishnaDN]

There are opensource implementation of this paper. Please check

[@akshay_pachaar]

Search for papers with code.

[@darkmythos4457]

very helpful, thanks for taking the time

[@ansupbabu8557]

very good channel..i am surprised to see ,less subscribers.

[@KrishnaDN]

Thank you. Hopefully I get more subscribers in future 😛

[@DamnightSC2]

good video, thank you :) also 4k :O

[@sehaba9531]

Thank you so much for this amazing explanation!

[@ruanjiayang]

Seq2Seq model is able to handle dimension variation in both input and outputs, which is one of the most implicit benefit of this series of models.

[@BalaguruGupta]

A well explained video. Thanks a lot!

[@shivaranashing7158]

Is it working on fractional input as well ?

[@tempdeltavalue]

Why do you ask?

[@8sukanya8]

Excellent explanation of the pointer networks! You are doing a huge favour to help us learn. I cannot help asking if you also know, how were problems containing variable outputs being handled before pointer networks were created? I mean how were variable length outputs cast as a fixed length outputs.

[@KrishnaDN]

If I understand your question , here is the answer from my perspective. One simple way used be just pooling in time . For example there are statistics pooling we use in speech which pools variable length features into a single feature representation. Other way could be to use state machine concept like Markov model assumption or using last time steps Hidden activation as the fixed dimensional feature for any variable length sequence.

[@vinitrinh]

Excellent video sir

[@varungarg8484]

Great job in making the video. The videos are really helpful

[@KrishnaDN]

Glad you like them!

[@cedrix57]

Thanks for this video. Is it possible to us pointer networks for 2D image to 3D model?

[@shivaranashing7158]

Did u get ans of this question?

[@cedrix57]

@@shivaranashing7158 No, I am waiting for a french class from a deep learning engineer in France. Until then, I am working on other projects.

[@tempdeltavalue]

🤯