Angehefteter Tweet
MTL DATA
861 posts
MTL DATA
@mtldata
Connecting nodes in the Montreal data community. Since 2014. #AIFest
Montreal, QC Beigetreten Mayıs 2014
461 Folgt530 Follower
MTL DATA retweetet
Thirteen critically important questions will be addressed by the new portfolio of CIFAR research programs:
cifar.ca/spring-2019-re…
#CIFARPrograms
English
MTL DATA retweetet
The 2019 Global AI Talent Report is out on my blog now. Demand grows and supply remains scarce. jfgagne.ai/talent-2019/
English
MTL DATA retweetet
MTL DATA retweetet
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.
ruder.io/thesis/
English
MTL DATA retweetet
Hand-labeling training data for machine learning problems is effective, but very labor and time intensive. This work explores how to use algorithmic labeling systems relying on other sources of knowledge that can provide many more labels but which are noisy.
Google AI@GoogleAI
In collaboration with @Stanford and @BrownUniversity, we present Snorkel Drybell, a variant of the open source Snorkel framework, which explores how existing organizational knowledge can be used as weak supervision to quickly label large training datasets. goo.gl/K2hKxk
English
MTL DATA retweetet
Excited about #ML and #Fashion ? 👚 Come to our @WiMLDS_Montreal meetup @element_ai on March 8th, to hear @negar_rz talk! Subscribe here: meetup.com/fr-FR/wimlds-m… 😃
#InternationalWomensDay #WiDS2019
English
MTL DATA retweetet
I originally thought of GANs as an unsupervised learning algorithm, but so far, to create recognizable object categories, they've needed a supervision signal / labeled images. This new work shows how to get them to work well with few labels.
twitter.com/MarioLucic_/st…
Mario Lucic@MarioLucic_
How to train SOTA high-fidelity conditional GANs usin 10x fewer labels? Using self-supervision and semi-supervision! Check out our latest work at goo.gl/idWNVs @GoogleAI @ETHZurich @TheMarvinRitter @mtschannen @XiaohuaZhai @OlivierBachem @sylvain_gelly
English
MTL DATA retweetet
The complaints "Python is slow" or "Python is unsafe" seem misguided.
The point of Python isn't to be fast or safe, it's to be flexible and hackable, and to interface well with everything else. It has become successful by serving as a frontend from which to call other libraries.
English
MTL DATA retweetet
MTL DATA retweetet
Last tweet for me on the OpenAI GPT-2 thing, but for those interested I think this video really elevates the discussion. Great work by all parties involved.
Sam Charrington@samcharrington
Really great discussion this eve on @OpenAI's recent language model release and the issues and considerations raised. Thanks @AmandaAskell @AnimaAnandkumar @Miles_Brundage @smerity @WWRob for participating. Those who missed it live can catch the replay at youtube.com/watch?v=LWDbAo…
English
MTL DATA retweetet
Wide Neural Networks of Any Depth Evolve as Linear Models Under Gradient Descent arxiv.org/abs/1902.06720 <--- this should blow your mind a bit!! Also holds for convolutional networks, batch norm, ... Also, closed form for test predictions resulting from gradient descent training.
English
MTL DATA retweetet
This is literally the first time I've seen am NLP researcher say they want to focus on helping normal people solve normal problems.
Hopefully the first of many :)
Lana Teplitsky@STeplitsky
Fascinating presentation by @yoavgo who is focusing his research on what people want: Turning text into structured data. “Information Extraction can be transformative to science” Excel for #NLP. #NaturalLanguageProcessing #AI @allen_ai BIU's NLP lab u.cs.biu.ac.il/~nlp/
English
MTL DATA retweetet
“As someone who is deeply interested in AGI, I find ImageNet much less interesting now, precisely because it can be solved with models that have little global understanding of images.”
Eric Jang@ericjang11
There is the neat and counterintuitive BagNet paper that is making the rounds on #ML Twitter. Here are my (quick) thoughts on it: blog.evjang.com/2019/02/bagnet…
English
MTL DATA retweetet
@keremturgutlu/understanding-building-blocks-of-ulmfit-818d3775325b" target="_blank" rel="nofollow noopener">medium.com/@keremturgutlu…
Here, I tried to explain building blocks of SOTA ULMFIT model. What is an AWD-LSTM? How Dropout is used everywhere? What is a QRNN and why might it be better? ...I also used excel spreadsheets to simplify things in a different way :)
English
MTL DATA retweetet
TIL when preparing my second deep learning lecture that the operators' manual for Rosenblatt's Mark 1 Perceptron machine was a classified document. It became unclassified only in 1977. The manual can now be found at apps.dtic.mil/dtic/tr/fullte…
English
MTL DATA retweetet
.@kaushal316 wrote a nice step-by-step tutorial on how to finetune BERT on a classification task (@kaggle Toxic Challenge)
Covers everything from data processing to model modification
Results are top-10% w. a very simple 30-lines-of-code single model 👇
medium.com/huggingface/mu…
English
MTL DATA retweetet
Such an inspiration: @alexandrecadrin is both a radiologist, and a kaggle-winning deep learning expert. He is the only one in the world so far - but there will be more to come!
Thank you @bhutanisanyam1 for helping to tell his amazing story.
hackernoon.com/interview-with…
English
MTL DATA retweetet
This Fri 2/8 is deadline to apply to fullstackdeeplearning.com taking place March 2-3 in Berkeley.
Also: @l2k from Weights & Biases / FigureEight and @jeremyphoward from fast.ai joined our guest speakers @RichardSocher (Salesforce) & Raquel Urtasun (Uber/Toronto).
English
MTL DATA retweetet
Thanks to @bhutanisanyam1 and @hortonhearsafoo, you can now run all the whole course.fast.ai lessons for free using Kaggle kernels.
course.fast.ai/start_kaggle.h…
English