NIPS 2018 Expo Workshop

Dec. 2, 2018

Expo Schedule »

Research to Production in NLP Applications at Facebook

Sponsor: Facebook, Inc

Abstract:

We propose a half day workshop on NLP at Facebook, critical component to many systems and applications. We will have an introductory talk on NLP efforts at Facebook, followed by more in depth talks about particular projects, a panel on Facebook's open source philosophy, and a tutorial on developing an NLP application from research to production.

Talk: Overview of NLP

This talk will give an overview of NLP efforts at Facebook and demonstrate how NLP is a critical component to many systems, ultimately to bring a better user experience for people on Facebook.

Talk: Fairseq

This talk will showcase NLP research conducted at our artificial intelligence lab and demonstrate how Fairseq, a general purpose sequence-to-sequence library, can be used in many applications, including (unsupervised) translation, summarization and dialog.

Talk: Translate

This talk will describe how PyTorch 1.0 allows researchers and engineer to move from research to production faster and describe the capabilities of Translate, our translation system engine, including models, export capabilities and low precision operators.

Talk: PyText

This talk will describe how PyText powers many NLP use cases at Facebook and how it is becoming a flexible NLP platform for both research and production use cases.

Panel: Open Source

Facebook researchers and engineers answer the following questions: How do you decide to open source a project? What steps are involved in releasing a project? What is the audience that you are targeting when releasing a project? How do you deal with maintenance and external requests from the community? What are the tangible benefits that you have observed by releasing projects so far?

Tutorial: Using PyTorch 1.0 Hybrid Frontend to Train and Export a seq2seq model

Hands-on tutorial on how to use PyTorch 1.0 for NLP; learn how to develop and train a sequence-to-sequence model and what steps are involved in making that model production ready.