Day 33: Microsft Research Open Data

Generative design —- this could be promising, and certainly fun https://t.co/mFoUT1LA1P

— Nando de Freitas (@NandoDF) June 21, 2018

ImageNet is hierarchical! When classifying images, we can show not just the leaves but the relevant subset of the entire tree. https://t.co/sCkJG9RaS1

— Mike Bostock (@mbostock) June 21, 2018

I analyzed 100k happy moments to find out what makes people the most happy, and it made me happy https://t.co/gZ1RUTk5KW pic.twitter.com/BRKbp9oy8u

— Nathan Yau (@flowingdata) June 21, 2018

How can you teach a machine learning system with human language rather than “labels”? With a semantic parser & labeling functions! New #ACL2018 paper by @bradenjhancock @paroma_varma @stephtwang @bringmartino @percyliang & Chris Ré @HazyResearch https://t.co/HMTdY5i5TZ #NLProc pic.twitter.com/VUsUdF2RjM

— Stanford NLP Group (@stanfordnlp) June 21, 2018

New #ACL2018 paper on StructVAE, semi-supervised learning with structured latent variables: https://t.co/4vQieQk91p
A new shift-reduce method for seq2tree neural models, semantic parsing results robust to small data, and nice analysis of why semi-supervised learning works! pic.twitter.com/9TKSkctOzC

— Graham Neubig (@gneubig) June 21, 2018

GLoMo: Unsupervisedly Learned Relational Graphs as Transferable Representations—learning a dependency graph to do deep transfer learning. Jake Zhao: “Perhaps this can also be seen as encoding some relational inductive bias into the machinery” 🤔 https://t.co/TkN7y3XDQb HT @ylecun

— Stanford NLP Group (@stanfordnlp) June 22, 2018

We won the Best Paper Award at the #CVPR2018 Deep Learning for Visual SLAM workshop (https://t.co/QhujldJKMY) for our paper on
Global Pose Estimation with an Attention-based Recurrent Network: https://t.co/H1lyIZahbb!! Congratulations to my co-authors Emilio, Jian and @rsalakhu. https://t.co/4OSny3lX05

— Devendra Chaplot (@dchaplot) June 21, 2018

Non-local Neural Networks: https://t.co/pu49vycxp9 #deeplearning #computervision pic.twitter.com/AUPZOc4qSV

— Learn PyTorch (@learnpytorch) June 21, 2018

Learning Intrinsic Image Decomposition from Watching Videos https://t.co/CfAKDumeo5

— Nando de Freitas (@NandoDF) June 21, 2018

In our most recent collaboration with Google Brain, we measure the similarity between neural network representations to provide insights into generalisation and the training dynamics of RNNs. @arimorcos @maithra_raghu

Blog: https://t.co/7qXQFkBMKI
Paper: https://t.co/G9ErnfJruG

— DeepMind (@DeepMindAI) June 21, 2018

Very excited about our latest preprint: https://t.co/ViXaxMq8RZ, joint work with @arimorcos and Samy Bengio. We apply Canonical Correlation (CCA) to study the representational similarity between memorizing and generalizing networks, and also examine the training dynamics of RNNs. https://t.co/5QmRxml0in

— Maithra Raghu (@maithra_raghu) June 21, 2018

The recent paper out from Google, "Scalable and accurate deep learning with electronic health records", has an notable result in the supplement: regularized logistic regression essentially performs just as well as Deep Netshttps://t.co/2vYzZiBoWRhttps://t.co/IStdZQOAe0 pic.twitter.com/U2qWwCb63p

— Uri Shalit (@ShalitUri) June 20, 2018

“Predicting the next Fibonacci number with Linear Regression in TensorFlow.js” by @curiousily https://t.co/FjrKK3UMph #tensorflow #deeplearning #machinelearning

— TensorFlow Practices (@TFBestPractices) June 21, 2018

#TransferLearning is crucial for general #AI, and understanding what transfers to what is crucial for #TransferLearning. Taskonomy (#CVPR18 oral) is one step towards understanding transferability among #perception tasks. Live demo and more: https://t.co/jGJSA4oIzM pic.twitter.com/Fl2UxyHYbl

— Berkeley AI Research (@berkeley_ai) June 6, 2018

Here is how we @ToyotaResearch do awesome large scale distributed #DeepLearning with @PyTorch in the Cloud for #AutomatedDriving with @awscloud 🤖🚘⛅️(we're hiring 😉): https://t.co/JvA0yPioyH

— Adrien Gaidon (@adnothing) June 21, 2018

Must-read for anyone who:
a. is living in reality
b. has no Scrooge McDuckian data-analysis vault…
"The Role of Resources in Data Analysis" by @rdpenghttps://t.co/a2OsmOqXtQ [🦆💰 animation by tedjohanssonsweden] pic.twitter.com/HvvXmxkfGE

— Mara Averick (@dataandme) June 21, 2018

Learning Structural Node Embeddings via Diffusion Wavelets. https://t.co/sjHyn4FJMu pic.twitter.com/Y3sNa6L3wB

— Jure Leskovec (@jure) June 14, 2018

How do you plot 3000 time series? Extract their features and plot the distribution! @robjhyndman #nyhackr pic.twitter.com/j4gcmE6eJe

— Emily Robinson (@robinson_es) June 21, 2018

Reproducibility tip of the day: If you're sharing code and data for people to reproduce your work, please release them both under a license that allows reuse. (I'm personally partial to Apache 2.0 and CC-BY-SA.) Unlicensed code + data is protected by copyright & not reusable. 😢

— Dr. Rachael Tatman (@rctatman) June 21, 2018

😍 sketchy *and* magnified, my #dataviz cup runneth over!
"ggrough: Convert ggplot2 charts to roughjs" 🖍 @xvrdmhttps://t.co/CJz6J5ryqk #rstats #ggplot2 pic.twitter.com/r4B2kVhpld

— Mara Averick (@dataandme) June 21, 2018

🚀 your team ⇨ R (+ good learning advice)...
"Teaching the tidyverse to co-workers" ✍️ @astroeringrand https://t.co/UA9YGdvWIt #rstats pic.twitter.com/IsZlBKDQr4

— Mara Averick (@dataandme) June 22, 2018

🌟 post for understanding Imports vs Depends:
🔭 "How R Searches and Finds Stuff" by @surajgupta
🔗 https://t.co/TIpb9OiJZy #rstats pic.twitter.com/rrrTiJLteD

— Mara Averick (@dataandme) June 21, 2018

Answering the question, What predictors are more important?, going beyond p-value thresholding and ranking https://t.co/IOr29G1T8a

— Andrew Gelman (@StatModeling) June 21, 2018

Here's a sneak peek at our new Federated Learning interfaces using @PyTorch and PySyft.

Try Federated Learning Using OpenMined: https://t.co/bSMgUZjgL1

Come Join our Slack: https://t.co/e1avrEWtIo pic.twitter.com/sHeQ2568gs

— OpenMined (@openminedorg) June 21, 2018

Check out Distiller, our @PyTorch based package for neural network compression research at https://t.co/px4g8yCrjS https://t.co/Fews6mK3X6

— Gal Novik (@gal_novik) June 22, 2018

Check out the #rstats fable package from @robjhyndman, a replacement for forecast! Many improvements including integrating with tidyverse packages #nyhackr pic.twitter.com/59rHAbRX78

— Emily Robinson (@robinson_es) June 21, 2018

A serious bug in #doc2vec fixed, after 3 years :O https://t.co/KBlA0XbLru
New release will have faster convergence & better vectors. Huge thanks to Umang!

— Gensim (@gensim_py) June 20, 2018

We're excited to announce the launch of Microsoft Research Open Data! This single, cloud-hosted location offers datasets representing many years of data curation and research efforts by Microsoft. Learn how it works: https://t.co/f4LhMxWlKF

— Microsoft Research (@MSFTResearch) June 21, 2018

On our blog, @stanford research engineer, @vsoch, makes a strong case for incentivizing public sharing and collaboration amongst academic researchers + lays out the challenges to making this happen. We hope Kaggle can contribute. Read the article here: https://t.co/U6ZOoCujcH

— Kaggle (@kaggle) June 21, 2018

New blog post in which I share my thoughts on the ICRA conference & state of robotics research! https://t.co/Qde92HjXF0

— Eric Jang (@ericjang11) June 22, 2018

Some say ML today is like alchemy more than science. That might be too harsh, but out of all the sciences, ML today is a lot like protein crystallography: you know when it works because you can measure it, but you have absolutely no idea what you need to do to make it work. https://t.co/UbFqgGJDHN

— David Pfau (@pfau) June 21, 2018

Investigating Human Priors for Playing Video Games https://t.co/OOPlV70fkR

— Nando de Freitas (@NandoDF) June 21, 2018

This article echoes my thoughts on AI - deep learning alone will not get us to AGI. Also what @GaryMarcus has been saying all along!https://t.co/hVlfz3XMZL

— Arun Shroff (@arunshroff) June 22, 2018

AI - the story so far:
1. Manually encoding all the knowledge you need for AI is hopeless.
2. Purely empirical learning methods keep exceeding expectations, but have limitations.
3. Solution: A little knowledge + A lot of data. How to do this: https://t.co/YaZ0pmwA60

— Pedro Domingos (@pmddomingos) June 21, 2018

Employee activism keeps spreading in tech—now Amazon employees are calling on the company to stop selling Rekognition to law enforcement and to boot Palantir from AWS https://t.co/6iJAQv00ub

— kate conger (@kateconger) June 22, 2018

I had no idea the tablets you sometimes find at chain restaurants are causing employees great trouble and threatening their livelihoods. Great @ceodonovan read here: https://t.co/62d2MRRlpr

— Tony Romm (@TonyRomm) June 21, 2018