search

colah / colah.github.io

colah.github.io
811 stars 191 forks source link

Confusing sentences for me in "Deep Learning, NLP, and Representations" #38

Closed dgkim5360 closed 6 years ago

dgkim5360 commented 6 years ago

Hello, Mr. Olah.

First thing first, thank you for your great work. I am learning with pleasure via your articles.

Here I would like to address what is difficult for me to understand. The below two sentences are from the source of the section "Shared Representations":

This general tactic ... One of the great strengths of this approach is that it allows the representation to learn from more than one kind of data.

There’s a counterpart to this trick. Instead of learning a way to represent one kind of data and using it to perform multiple kinds of tasks, we can learn a way to map multiple kinds of data into a single representation!

It was kind of confusing for me, because "the trick" and its "counterpart" seem to have the similar descriptions. And below is my current understanding.

For the first one, the representation can learn other kind of data, incrementally, whenever other tasks come in and the word embedding tries to solve the tasks. And for the second one (its counterpart), assuming we already have multiple tasks and its corresponding multiple kinds of data, then the single representation can learn these all data.

Since I have worked this article for the translation in Korean, it would be great if my understanding gets confirmed and corrected.

Best regards, Don

colah commented 6 years ago

Hi Don!

The first one is about using a single representation for multiple tasks:

This general tactic – learning a good representation on a task A and then using it on a task B – is one of the major tricks in the Deep Learning toolbox.

The second one is about embedding multiple kinds of data into a single represetnation

Instead of learning a way to represent one kind of data and using it to perform multiple kinds of tasks, we can learn a way to map multiple kinds of data into a single representation!

dgkim5360 commented 6 years ago

Yep, now I understand. Thank you for the gist ! :)

I will close this issue.

FYI, I finished the correction in my translation, which you can check at the below link. http://dgkim5360.tistory.com/entry/deep-learning-nlp-and-representations-kr