Human Performance on SQuAD2.0 has now been surpassed! Been in the making for the last year, but finally achieved! Congratulations Joint Laboratory of HIT and iFLYTEK Research on this milestone! @KCrosner @stanfordnlp pic.twitter.com/MyVU3faDhL
— Pranav Rajpurkar (@pranavrajpurkar) March 27, 2019
"SciBERT: Pretrained Contextualized Embeddings for Scientific Text," Beltagy et al.: https://t.co/WNR8pR99y7
— Miles Brundage (@Miles_Brundage) March 27, 2019
Current methods for debiasing word embeddings are insufficient. They hide bias, rather than removing it, and a lot of the bias information can be recovered. https://t.co/s7RUtYDSnX pic.twitter.com/Q9h71vsnof
— Rachel Thomas (@math_rachel) March 25, 2019
Visualizing memorization in RNNs — A new Distill article by @andreas_madsen.https://t.co/urqxBDxiME
— distillpub (@distillpub) March 25, 2019
🎶 @HalfEatenScone's songsim algo ggplot2-ified by @pprevos!
— Mara Averick (@dataandme) March 25, 2019
"Decode Lyrics in Pop Music: Visualise Prose with the Songsim algorithm" https://t.co/r5xBXT61QO #rstats #dataviz
/* Other than ABBA, song choices by yours truly 😜 */ pic.twitter.com/Jd7hTWXMAX
My PhD thesis Neural Transfer Learning for Natural Language Processing is now online. It includes a general review of transfer learning in NLP as well as new material that I hope will be useful to some.https://t.co/PxfVRYuyjx pic.twitter.com/XrYYYHx1ln
— Sebastian Ruder (@seb_ruder) March 23, 2019
Fine-tuning GPT-2-117M on jokes - off to a rocky start:
— Miles Brundage (@Miles_Brundage) March 23, 2019
"I just saw this guy with six eyes going to the restroom. He looks like a ghost."
"My girlfriend keeps asking me what I think of her 'soul' ... I do wonder how she can be so narcissistic."
Should be super easy to install & use (see the image above)
— Thomas Wolf (@Thom_Wolf) March 21, 2019
👉 pip install pytorch-pretrained-biggan
👉 https://t.co/L2RGZUWeIc
Implemented from the raw computational graph of the TF Hub module. Pretrained weights are the ones pre-trained by @ajmooch at DeepMind. Thanks Andrew!
I needed a good GAN to tweak for a CV+NLP project so here is a *pretrained* version of BigGAN in PyTorch
— Thomas Wolf (@Thom_Wolf) March 21, 2019
Sweet stuff:
-AFAIK code is not public yet so here is an op-for-op implem to read/tweak
-Checkpoints 2x smaller (no dead vars)
-Print images in terminal (icing on the cake)👇 pic.twitter.com/GbE7cDfVcX
https://t.co/9IHozCXzzc --- r/ML gets the small GPT2 to write imaginary news. I didn't expect the small model to be that coherent.
— Ilya Sutskever (@ilyasut) March 20, 2019
v2.0.18 was never supposed to be that slow — but if your numpy ended up linked to a bad BLAS kernel, performance could be pretty poor. Here's how we fixed this — it was a terrible ordeal and I'm glad it's over 🙃 https://t.co/Wm1X9UcwPS
— Matthew Honnibal (@honnibal) March 19, 2019
#GluonNLP 0.6 is out. Now with full Bert support. Full changelog is here: https://t.co/zP5hBM7TEs
— Alex Smola (@smolix) March 19, 2019
