Cool post by @hxiao, author of bert-as-a-service that highlights the increasing focus on pretrained models and end-to-end applications. It also describes a generic IR system that can scale to an arbitrary number of encoders. https://t.co/nh3FGoeYvS pic.twitter.com/0XnXbUrRnf
— Sebastian Ruder (@seb_ruder) August 8, 2019
"ViLBERT: Pretraining Task-Agnostic Visiolinguistic Representations for Vision-and-Language Tasks," Lu et al.: https://t.co/2xJgxhFLhC
— Miles Brundage (@Miles_Brundage) August 7, 2019
"Our work represents a shift ... towards treating visual grounding as a pretrainable and transferable capability."
#acl2019nlp paper on "Beyond BLEU: Training NMT with Semantic Similarity" by Wieting et al.: https://t.co/5N9SBiPyDq
— Graham Neubig (@gneubig) August 6, 2019
I like this because it shows 1) a nice use case for semantic similarity, 2) that we can/should optimize seq2seq models for something other than likelihood or BLEU! pic.twitter.com/Fh8WJe5tKH
The other advantage of QRNNs that I usually bring up is that a complex recurrence function such as the LSTM's isn't always an advantage. In the Transformer-XL paper they show that a QRNN is better at their context metric than an LSTM with the same parameter budget. pic.twitter.com/QtupPkwDWQ
— Smerity (@Smerity) August 6, 2019
You can add this one that focus on summarization https://t.co/F2W7LjXCtZ
— Thomas Scialom (@ThomasScialom) August 6, 2019
Only a weekend after @ACL2019_Italy, there are already awesome reviews available on various topics:
— Sebastian Ruder (@seb_ruder) August 5, 2019
- Trends in NLP by @mihail_eric https://t.co/w2onhvmBxx
- Knowledge graphs by @michael_galkin https://t.co/9wMPKLId38
- MT by @noecasashttps://t.co/7eMlbhHWrL
👏#acl2019nlp
Does maximizing information in an (artificial) neural network lead to useful representations? Maybe, sometimes, but network architecture can be more important https://t.co/Gs838LrMy6 pic.twitter.com/uSvAH8Z5Pu
— Adam J Calhoun (@neuroecology) August 5, 2019
spaCy meets PyTorch-Transformers: Fine-tune BERT, XLNet and GPT-2 · Blog · Explosion https://t.co/JSWsJgf5T0 via @rightrelevance thanks @peteskomoroch
— Bojan Tunguz (@tunguz) August 2, 2019
Dialogue research is crucial for building the next gen of intelligent #AI systems. We've made scientific advancements in 5 areas that open-domain chatbots fail in today: specificity, consistency, empathy, knowledgeability, and multimodal understanding. https://t.co/rg55v15G89 pic.twitter.com/aIc9nGtk3B
— Facebook AI (@facebookai) August 2, 2019
OpenKiwi is a framework for Machine Translation Quality Estimation from @Unbabel
— PyTorch (@PyTorch) August 2, 2019
It just won the Best Demo Paper award at @ACL2019_Italy , congrats to the team!
Blog Post: https://t.co/UqJLicBOsJ
Get Started: https://t.co/p7dg0dbSg5 pic.twitter.com/bwVwsZNexz
What Does BERT Look At? An Analysis of BERT's Attention. Kevin Clark @clark_kev, Urvashi Khandelwal @ukhndlwl, Omer Levy @omerlevy_, Christopher D. Manning @chrmanning. https://t.co/d0uweJ2VRq https://t.co/rJA7HmrAU4
— Stanford NLP Group (@stanfordnlp) August 2, 2019
Facebook AI wins the 2019 WMT competition in the four language directions it participated in.
— Yann LeCun (@ylecun) August 1, 2019
Blog post + paper + trained models are available. https://t.co/5TfK0hCK7A
