(Today is Memorial Day in the U.S., hence data science related tweets are sparser.)
Non-Autoregressive Machine Translation Model
“Theory and Experiments on Vector Quantized Autoencoders” https://t.co/yV12G9yUFf
— hardmaru (@hardmaru) May 29, 2018
We develop a non-autoregressive machine translation model whose accuracy almost matches a strong greedy autoregressive baseline Transformer, while being 3.3 times faster at inference. Joint work with @ashVaswani @nikiparmar09 Aurko Roy https://t.co/gtLPhMXFHh
— Arvind Neelakantan (@arvind_io) May 29, 2018
On Data
As deep learning progressed faster and I understood more of it, I've realized it changed my concept of what data is and how it will be treated. Data won't stay static. Data will be given its own compute and agency, allowing itself to iteratively refine understanding over time.
— Smerity (@Smerity) May 28, 2018
The same patterns done by humans manually today will become transformations on underlying data. When it's automated there won't be a clear delineation between software and data. A video can become a thumbnail or soundscape, the video processed for animals or faces, and ...
— Smerity (@Smerity) May 28, 2018
disparate knowledge woven in to fill details. If there was out of focus shots of a wild animal, video de-blurring may find similar wild animal images to "paint the gap". When this is all standard for everyone and as normal as typing on a keyboard, data will be damn near magical.
— Smerity (@Smerity) May 28, 2018
Think about how static people must have thought images were before the advent of digital photo editing (=film). In that situation the negative was still there but it was hard to extract value when there's a bottleneck of professionals. That's the state data is in right now.
— Smerity (@Smerity) May 28, 2018
Double-blind Review
Our article on the effectiveness of anonymization in double-blind review is out! TL;DR - it’s really effective, undermining a key argument against double-blind review. (Co-authored by @clegoues, Yuriy Brun, @SvenAppel, @YSmaragdakis, and Safraz Khurshid) https://t.co/VF3vq3WTd9
— Emery Berger (@emeryberger) May 24, 2018
OMG reviewers with author information were more likely to recommend acceptance of
— Sylvain Chabé-Ferret (@SylvainCF) May 25, 2018
- papers from famous authors (+76%)
- papers from top institutions (+67%)
- papers with male authors (+19%)
Horrible biases, same papers evaluated differently (probably guilty myself) https://t.co/xDtuq0SsuZ
Yellowbrick
I’ve seen Yellowbrick being mentioned several times. Probably worth trying out.
Yellowbrick: A #Python #dataviz package for #MachineLearning. Works well with #sklearn. Seems like a nice alternative to scikit-plot.https://t.co/h3U3rMKTho pic.twitter.com/fP3vZ24DnH
— Randy Olson (@randal_olson) May 28, 2018
Notables
Excited to share a new paper, which investigates how faculty hiring is a mechanism for how "Prestige drives epistemic inequality in the diffusion of scientific ideas," with @alliecmorgan D. Economou, and @samfway https://t.co/AWJgCTw7Hx pic.twitter.com/oa9gWx476x
— Aaron Clauset (@aaronclauset) May 28, 2018
"Empiricism and the limits of gradient descent"https://t.co/NPYxCCR81s
— Julian Togelius (@togelius) May 27, 2018
New blog post where I use Karl Popper's take on epistemology to argue that evolutionary algorithms can learn in a way that backpropagation cannot.
Why you need to improve your training data, and how to do it: https://t.co/ArVq78E6vX
— Pete Warden (@petewarden) May 28, 2018
Interested in learning about #NLP? We compiled some of our favorite (free) NLP & #machinelearningeducational resources here: https://t.co/DQLqRBtGer by @seb_ruder
— AYLIEN (@_aylien) May 27, 2018
Watch Carson Sievert talk about Creating Interactive Web Graphics for Exploratory Data Analysis. Interactive plots can be tricky to dev & less-popular exploratory analysis. But that's less and less so now@cpsievert @plotlygraphs https://t.co/1Nx1thMQeJ #rstudioconf #plotly pic.twitter.com/0y6BPU8WlR
— RStudio (@rstudio) May 28, 2018
Miscellaneous
How do your hobbies make you a better scientist?
— Science Magazine (@sciencemagazine) May 28, 2018
Share your thoughts by Friday—your answer could be published in a forthcoming issue of Science: https://t.co/eDzMwPAQEf #NextGenSci
Normal brain connectivity vs while tripping on magic mushrooms—recreated using needle and thread for #CraftyDataViz pic.twitter.com/RAgjMQYPfe
— Patti Shih (@pattithepotato) May 20, 2018
Turns out that healthcare is complicated. And combining medicine, math, and computing to produce breakthroughs is hard. Still someone will do it. Just not IBM right now. https://t.co/AM9CEnwjsn #AI #MachineLearning #BigData #DigitalHealth pic.twitter.com/eZccq5K3qG
— Harlan Krumholz (@hmkyale) May 28, 2018
That does it. I am officially switching back to “Statistician”. pic.twitter.com/qpHO1qXpzk
— David Smith (@revodavid) May 29, 2018
We need an internal theory of change when trying to influence management. Dilbert has useful advice on what works. https://t.co/4YGZsHjLvQ pic.twitter.com/r87nkfC1DZ
— Duncan Green (@fp2p) May 29, 2018
Five years on, it's disappointing that the data visualization community never fully recognized the benefits of 3D stacked scatter pie columns: https://t.co/egvMKQ7WbU pic.twitter.com/ay9Zngedrl
— Daniel MacArthur (@dgmacarthur) May 28, 2018
Amazing use of data+tech. Hats off to @codeforamerica https://t.co/cWg2eEbcEp
— dj patil (@dpatil) May 29, 2018
Just a robot riding a bicycle pic.twitter.com/TWSKheldld
— Vala Afshar (@ValaAfshar) May 27, 2018