FairSeq Toolkit

FairSeq Toolkit - Major Update
- Distributed Training
- Transformer models (big Transformer on WMT Eng-German in < 5 hours on DGX-1)
- Fast Inference: translations @ 92 sent/sec for big Transformer
- Story Generation
Read more at Michael Auli's post: https://t.co/eptKDuh0WI pic.twitter.com/d4OtJZpdFw

— PyTorch (@PyTorch) June 16, 2018

code and pre-trained models to reproduce the recent paper "Scaling Neural Machine Translation" (https://t.co/mrRDmlwax1) where we train on up to 128 GPUs with half precision floating point operations as well as delayed batching.

— PyTorch (@PyTorch) June 16, 2018

Create a ChatBot with tf-seq2seq

How To Create a ChatBot With tf-seq2seq For Free! – Deep Learning as I See It https://t.co/3UuHNI2gwV #AI #DeepLearning #MachineLearning #DataScience

— Mike Tamir, PhD (@MikeTamir) June 16, 2018

Visualization

Comparing values is a fundamental and critical task in any data analysis. And yet, so many subtleties! A visual meta-comparison. https://t.co/uDbeNCkNeZ

— Mike Bostock (@mbostock) June 15, 2018

These four maps show the same exact data, but look wildly different. What the? Here’s why: https://t.co/uDbeNCkNeZ pic.twitter.com/0vPQSg7qBw

— Mike Bostock (@mbostock) June 15, 2018

Outflow of #refugees from #Syria to other countries in 2016. #datavizhttps://t.co/n6TAWDdeFj pic.twitter.com/QcEATAGyuG

— Randy Olson (@randal_olson) June 15, 2018

Notable Research

While there are many attempts to map Auto ML to deep learning, to automate the tedious hyperparam tuning, here's a smart alt. via good initialization schemes: "Dynamical Isometry and a Mean Field Theory of CNNs: How to Train 10,000-Layer Vanilla ConvNets" https://t.co/zt9o89uP0K

— Sebastian Raschka (@rasbt) June 15, 2018

Improving Consistency-Based Semi-Supervised Learning with Weight Averaging
Our new paper (+code!): https://t.co/5dRLFqgJlh
By analyzing loss geometry, we achieve record semi-supervised results, including 95% accuracy on CIFAR-10 with only 4000 labels! pic.twitter.com/EfqDhr3qC3

— Andrew Gordon Wilson (@andrewgwils) June 15, 2018

Text to speech with few data. https://t.co/JiX6TglXAw

— Nando de Freitas (@NandoDF) June 15, 2018

Autoregressive Quantile Networks for Generative Modeling: https://t.co/MImdcFiYxj

— DeepMind (@DeepMindAI) June 15, 2018

Tutorials and Resources

This week's #KernelAwards winner uses the Stack Overflow 2018 Developer Survey to better understand what kinds of users are likely to identify as part of the Stack Overflow community: https://t.co/BMcvm7Pv2B pic.twitter.com/SGbu5BhLNu

— Kaggle (@kaggle) June 15, 2018

NCRF++ : A neural CRF++ toolkit for sequence labeling tasks. Works pretty much similar to Taku's CRF++ package, but built with #PyTorch! #nlproc https://t.co/05ZL6bEcoO pic.twitter.com/ME8k7ruMdA

— Delip Rao (@deliprao) June 16, 2018

GitLab's Web IDE looks pretty nice. Sooner or later entire workflows will take place in browser tabs. https://t.co/SBrBMfJb0x pic.twitter.com/Ts6OHuDVyk

— hardmaru (@hardmaru) June 15, 2018

Libraries like Vega-Lite and Vega, that are built on top of #D3js, really deserve to be better known. Unless the visualization you want to build is really novel, you'll be much better served by starting with those! @vega_vis #DataScience #dataviz https://t.co/isSS5eYJxk

— Christian Hudon (@christian_hudon) June 14, 2018

Python

Clean architectures in Python: a step-by-step example – https://t.co/sxye7fOSuh

— Pycoders Weekly (@pycoders) June 15, 2018

rstats

Save time with RStudio code snippets! See how they work -- and how to write your own -- in my latest "Do More With R" #rstats screencast: https://t.co/WtnqPsOKkE pic.twitter.com/bBW5wrmZzc

— Sharon Machlis (@sharon000) June 15, 2018

🌟 guide by @mjfrigaard:
"Getting started w/ stringr for textual analysis in R"https://t.co/ZX27YWn6BQ via @storybench #rstats #stringr pic.twitter.com/7834zO817y

— Mara Averick (@dataandme) June 15, 2018

new blog post: a casual case study on how to speed up #rstats code featuring xml2, rcpp and furrrhttps://t.co/KOACH1O8dy pic.twitter.com/63T0SMsxH7

— alex hayes (@alexpghayes) June 15, 2018

ICYMI, 😱 inputs, interfaces & more:
"The R Shiny 📦s you need for your web apps" by @AntGuilllot https://t.co/gtTZRGjr0v #rstats #rshiny pic.twitter.com/5fLF62qAJZ

— Mara Averick (@dataandme) June 16, 2018

🌟 guide by @mjfrigaard:
"Getting started w/ stringr for textual analysis in R"https://t.co/ZX27YWn6BQ via @storybench #rstats #stringr pic.twitter.com/7834zO817y

— Mara Averick (@dataandme) June 15, 2018

My #xaringan slides on how use #xaringan to make slides with #rstats: https://t.co/4lZudRoGBn Raw Rmd here: https://t.co/QeFEdYlaGv pic.twitter.com/7QOUqawFQf

— Dr. Alison Hill (@apreshill) June 13, 2018

Miscellaneous

We haven't yet solved even 10% of the problems we could solve with existing AI/ML techniques. Even if new research were to deliver nothing from now on, there still wouldn't be another AI winter. AI/ML will keep on delivering for years to come.

— François Chollet (@fchollet) June 16, 2018

Fortunately, this should improve the productivity and focus of our community, rather than harm it. Our biggest successes are still ahead

— François Chollet (@fchollet) June 16, 2018

—Deliberately design end-to-end workflows.
—Reduce cognitive load for your users.
—Provide helpful feedback to your users.

In the long run, good design always wins, because it makes its adepts more productive and impactful.

Good design is infectious. ✨https://t.co/KdG7IT7Kwh pic.twitter.com/bXKkQpqJLP

— 👩‍💻 DynamicWebPaige @ GOTO; Amsterdam 🌷🇳🇱 (@DynamicWebPaige) June 15, 2018

Something you develop the longer you do data analysis is a "spidey sense" when something "looks wrong" that can often lead to uncovering a problem with your dataset or analysis. pic.twitter.com/7EwcXiVoyu

— Data Science Renee (@BecomingDataSci) June 15, 2018

@ceshine_en

Inpired by @WTFJHT