January 2019 - Bag of Tricks in Machine Learning

CV papers

January 31, 2019January 31, 2019 by admin

https://github.com/hoya012/deep_learning_object_detection

Ground truth?

January 31, 2019January 31, 2019 by admin

If you want to question whether human-produced labels for training ML models represent "ground truth," look no further than the sheer volume of misclassifications in a pair of recycling/trash bins.
— Sean J. Taylor (@seanjtaylor) January 29, 2019

To read more books and less paper

January 27, 2019January 27, 2019 by admin

CMU 10-708 PGM by Eric Xing

January 27, 2019January 27, 2019 by admin

https://www.youtube.com/channel/UCim-E6bNz7lUyKZwhgN6S1A/videos

How to Structuring Python Project

January 27, 2019January 27, 2019 by admin

https://docs.python-guide.org/writing/structure/

Fast ai course v3

January 26, 2019January 26, 2019 by admin

https://course.fast.ai/

Natural Questions Corpus (Google AI)

January 25, 2019January 25, 2019 by admin

https://ai.googleblog.com/2019/01/natural-questions-new-corpus-and.html

在文本分类任务中，有哪些论文中很少提及却对性能有重要影响的tricks？

January 24, 2019January 24, 2019 by admin

https://www.zhihu.com/question/265357659

把Cross Entropy梯度分布拉‘平’ – AAAI 2019 Oral

January 23, 2019January 23, 2019 by admin

AAAI 2019 Oral | 把Cross Entropy梯度分布拉‘平’，就能轻松超越Focal Loss
https://www.jiqizhixin.com/articles/2019-01-17-30

Kaggle Mercari Price Suggestion Challenge (1 place) — Pawel Jankiewicz, Konstantin Lopuhin

January 22, 2019January 22, 2019 by admin

ShujianFollow

Retweet on Twitter Shujian Retweeted

Aran Komatsuzaki @arankomatsuzaki·

Nick Ryder (an author of GPT-3) gives a talk on GPT-3, in which he discusses a lot of things not written in the paper. W&B talks like this are highly underrated with full of great MLers. Don't forget to check other talks, too.

https://t.co/CcWWB6iO1e

Reply on Twitter 1291539556588756995 Retweet on Twitter 12915395565887569959 Like on Twitter 129153955658875699540 Twitter 1291539556588756995

Retweet on Twitter Shujian Retweeted

Emanuele Rossi @emaros96·

The code for our recent paper "TGN: Temporal Graph Networks" is live on Github: https://t.co/5NZPOZnPJ9 !
Paper link: https://t.co/1gUTxlC3pr
Joint work w/ @mmbronstein @frederickmonti @b_p_chamberlain @ffabffrasca @aittalam

Reply on Twitter 1290941116880224257 Retweet on Twitter 129094111688022425731 Like on Twitter 1290941116880224257145 Twitter 1290941116880224257

Retweet on Twitter Shujian Retweeted

Amanda Askell @AmandaAskell·

The internet: We have no idea how good GPT-3 is because all of the outputs are cherry-picked.

Me, who made sure the human evaluation experiments in the paper didn't involve any cherry-picking: 😢

Reply on Twitter 1291119356844183552 Retweet on Twitter 129111935684418355237 Like on Twitter 1291119356844183552639 Twitter 1291119356844183552

Retweet on Twitter Shujian Retweeted

Suraj Patil @psuraj28·

ONNXT5: accelerated inference for T5, built using
@onnxruntime and @huggingface transformers.
Run any of the T5 trained tasks (translation, summarization, sentiment analysis, completion, generation) at blazing speed 🔥. Built by Abel
https://t.co/m18PjNHFug

Reply on Twitter 1291044115937812481 Retweet on Twitter 129104411593781248130 Like on Twitter 1291044115937812481107 Twitter 1291044115937812481

Retweet on Twitter Shujian Retweeted

Leo Boytsov @srchvrs·

"Deep metric learning papers have consistently claimed great advances in acc. We take closer look to see if it is actually true. We find flaws in experimental methodology of numerous metric learning papers & show that actual improvements over time have been marginal at best." https://t.co/cTOPgpmg9x

Reply on Twitter 1291143678522798080 Retweet on Twitter 12911436785227980802 Like on Twitter 129114367852279808014 Twitter 1291143678522798080