Deep learning

[patperry]

What are the recent developments in deep learning based methods for analyzing text? What software makes these methods feasible? What are the important implementation decisions? What are the limitations of these tools? What needs to change to enable the next set of developments?

[goodmansasha]

I’d like to attend this session

[jwijffels]

It would be nice to have an R wrapper around dynet

[sleepinyourhat]

I might join, as the friendly DL-for-NLP researcher from down the hall.

[adamobeng]

I'm also interested

[vanatteveldt]

I'd like to attend too (as an ignorant DL-newb)

[trinker]

Interested, ignorant newb status as well

[bmschmidt]

Interested as well. Interested in talking about pre-trained sentence embeddings: https://www.tensorflow.org/hub/modules/google/universal-sentence-encoder/1

[trinker]

@bmschmidt I am unsure how word embeddings are useful in a practical aplication, i.e., how can I use them in modeling? I'd love to hear about this from knowledgeable folks.

[haven-jeon]

I want to attend.

It's difficult to include deep learning model into R or Python package and make it works on any envirements. One of my package KoSpacing works well on test envirement, but not sure can be work on other envirement. It's very heavy and depends many other package like tensorflow(gpu, cpu....), keras. It needs to be more lighter and lighter.

Any idea? Can this help? : https://github.com/harmanpreet93/keras-model-to-cpp

[amatsuo]

Interested in it as well.

[kyunghyuncho]

i want to attend.

[kyunghyuncho]

@haven-jeon this is related to the interoperability (https://github.com/ropensci/textworkshop18/issues/8). i feel like it would be a good idea to have some kind of common interface for inference. once the interface is standardized (or informally agreed upon by a few), we could use e.g. zeromq to call an inference function implemented in another framework from R or others.

[haven-jeon]

@kyunghyuncho It will be good to have common inference interface like zeromq as you mention.