Natural Language Processing: The Age of Transformers

Natural Language Processing: The Age of Transformers(blog.scaleway.com)

200 points by ilnmtlbnm 6 years ago | 24 comments

baalimago 6 years ago |

>Next we shall take a moment to remember the fallen heros, without whom we would not be where we are today. I am, of course, referring to the RNNs - Recurrent Neural Networks, a concept that became almost synonymous with NLP in the deep learning field.

XLNet (https://arxiv.org/abs/1906.08237) is in essence a recurrent neural network, using a transformer (which is based on neural networks) which recurrently keeps context between different batches. But the gated RNN's, such as AWD-LSTM/GRU, are fading out to the superior transformer architectures, this is true.

That's my only complain though, excellent theoretical introduction.

Although, if anyone wanted to actually implement a transformer, be ware that you want to have a 8+ GB GPU unit available, or be prepared to use cloud computing (Google Colab is free, for now). Training neural networks is quite hardware dependent still.

karambahh 6 years ago | |

Scaleway (where the author of this post works, as I do) is a cloud service provider with a pretty interesting GPU instance: Nvidia P100 16-GB NVIDIA Tesla P100 at 1€ per hour

bitL 6 years ago | |

RNNs are still useful in actual time-dependent sequences like activity detection, self-driving car steering etc. though even those are getting enhanced by using attention; use of RNNs in NLP was more of a necessity as there were no other Deep Learning models capable of delivering some results on arguably sequential nature of NLP (let's say that is a quite imperfect assumption). As attention allows viewing the whole input at once, it's easier for non-linear optimizer to set meaningful weights without getting into recursion, though that comes at massive memory cost (i.e. forget about using 2080Ti for NLP).

abhgh 6 years ago | |

I was going to mention XLNet before I saw your comment.

Also, a recent piece of interesting work [1] shows that with the right control parameters, you could still use gated RNNs, like LSTMs, for pretty good language modeling.

[1] http://www.abigailsee.com/2019/08/13/what-makes-a-good-conve...

sgt101 6 years ago |

If anyone wants to use these tools practically I urge you to have a good look at this paper : https://www.aclweb.org/anthology/P19-1439/

My take away - pretraining achieves excellent paper results but robust application is hard. There is still quite a way to go down this road for fault intolerant users and applications.

Jack000 6 years ago | |

I'm not an expert, but after playing with some pre-trained transformers I think they are mostly good at the exact thing they're trained for. eg. GPT-2 is great for text generation, but if you try to use it for say translation, it will tend to add imagined details not in the source text. Similarly, BERT is great at sequence and token-level classification but quite bad at text generation.

bra-ket 6 years ago |

Previous discussion on “Attention is all you need”: https://news.ycombinator.com/item?id=15938082

Recent work by Jakob and team:https://ai.google/research/people/author37567/

Image transformer is particularly interesting

cjauvin 6 years ago | |

Also, recent discussion about a very good intro article about Transformers: https://news.ycombinator.com/item?id=20773992

kuu 6 years ago | | |

I was about to share it too, really GREAT intro :)

macando 6 years ago |

"This spring, the Elon-Musk-founded AI research lab OpenAI made a splash with an AI system that generates text. It can write convincing fake reviews, fake news articles, and even poetry."

You can check it out for yourself. https://talktotransformer.com/ Pretty awesome.

mark_l_watson 6 years ago |

Good explanation of transformers, and the history leading up to them. I look forward to the next installment covering BERT.

As someone who spent a lot of time trying to manually code up solutions to anaphora resolution (pronoun coreference), BERT seemed like a small miracle to me. As a side comment: I love that getting training data for BERT is so cheap: any text source, and randomly remove words, target output is predicting the words removed.

ArtWomb 6 years ago |

Conversational AI is much closer than we think. Neural sequence-to-sequence models are successful in domain specific domains. But in the context of chit-chat based dialogue systems, the responses lack humanity. Undoubtedly due to the fact they don't comprehend our world. Transfer learning alleviates some of that awkwardness.

If anyone's interested in further experiments on their own. There is now a unified Python framework for dialogue models ;)

https://parl.ai/about/

abhishek0318 6 years ago | |

Really? Most of the chatbots in production are rule based systems.

blurbleblurble 6 years ago | | |

The transformer can also be used at different levels of abstraction, e.g. to do interesting stuff with knowledge graphs. I think the transformer architecture is about to make things very, very interesting.

https://arxiv.org/pdf/1904.02342.pdf

p1esk 6 years ago | |

We don’t even know how many breakthroughs it will take to get to the level of 3yo’s conversational abilities. And I’m talking major level breakthroughs, like LSTM or Transformer, probably trained on much more data.

fellahst 6 years ago |

Great blog!