tensorflow/neural-structured-learning

Possible for NSL to be use in Seq2seq setting?

joelowj opened this issue · 2 comments

joelowj commented

Hi Team,

I chanced upon NSL framework and would like to say that this is an amazing work. I am currently working on some Structured Prediction task and would like to explore this framework. But I am not too clear if is is possible for this framework can be extended to say a Seq2Seq problem? Based on your experience and familiarity is it possible to do so? At the moment, is it possible for me to pack more than 1 graph and augment it with the training set?

DualityGap commented

Thanks for the question, Joel : )

Yes, it is possible to extend this framework to Seq2Seq problems.
For example, see Figure 1(D) in this paper.
A graph can be used to regularize the output of an encoder, and then this encoder output (embedding) will be sent to the decoder in a Seq2Seq setting.

Regarding your 2nd question, packing more than one graph may require some light-weight modifications on pack neighbors to incorporate more than one graph : )

joelowj commented

@DualityGap cool, thanks a lot for the clarification. Have a nice day!