Today we witness the birth of the world’s first psychopath AI - Norman. And another beutiful work by Nathan Yao @flowingdata.
Spiking Neural Networks
🎉🎉🎉 I'm excited to announce the release of our spiking neural networks (SNNs) simulation library, BindsNET! The code is all written in Python using @PyTorch CPU and GPU n-dimensional Tensors.
— Dan Saunders (@djsaunde) June 6, 2018
code: https://t.co/QpAu5jUsC0
pre-print: https://t.co/RTBK8lIwm7
Psychopath AI
MIT scientists created a "psychopath" AI by feeding it violent content from Reddit https://t.co/ZJ83rfW82H pic.twitter.com/Lx6nBxmwq7
— The A.V. Club (@TheAVClub) June 7, 2018
Visualization
Saving time and gaining flexibility while working remotely https://t.co/9mbPjXcOY2 pic.twitter.com/D1egeyWKFM
— Nathan Yau (@flowingdata) June 6, 2018
I collected NYC subway countdown clock data every minute for five months and wrote about it here: https://t.co/vwJWutGF4u pic.twitter.com/99RGV50LAj
— Todd Schneider (@todd_schneider) June 6, 2018
Still my favorite visual of why PowerPoint is a less powerful tool for communicating than Word. Love @EdwardTufte and his booklet ‘The Cognitive Style of PowerPoint: Pitching Out Corrupts Within’ (2006). And he’s right, PPT presentations do emulate the Dick and Jane book format. pic.twitter.com/ehS3UacV0k
— (((Rob Nelson))) (@RobLeeNelson) June 6, 2018
8 US State populations fit into NYC@MetricMaps #demographics #ElectoralCollege pic.twitter.com/YLzSnOhmHT
— Mark Abraham (@urbandata) November 15, 2017
Notable Research
Interested in AI and cognition?—read @GaryMarcus's 'Deep Learning: A Critical Appraisal! Crystal-clear summary of current challenges in #DeepLearning, with a positive role to be played by cognitive science going forward. Agree with so much here https://t.co/TMSteZCjNK
— Courtney Hilton (@courtneybhilton) June 6, 2018
Just finished reading through DeepMind's / @PeterWBattaglia et al.'s new review/position paper on graph neural nets: https://t.co/BtiATZg4jB Timely and highly relevant contribution to the field IMO, couldn't agree more with their motivation for this class of models
— Thomas Kipf (@thomaskipf) June 5, 2018
Honored to receive the best paper award in COLT: "Algorithmic Regularization in Over-parameterized Matrix Sensing and Neural Networks with Quadratic Activations" https://t.co/7gXgwdOcQ1. Congrats to Yuanzhi and Hongyang!
— Tengyu Ma (@tengyuma) June 6, 2018
When the segmentation tool works so well that it almost directly gives you layers. Nice paper from @ofgulban: https://t.co/urrEE2Ll6z pic.twitter.com/NYORWQ95jb
— layerfMRI (@layerfMRI) June 6, 2018
"Deep Video Networks" tee-up a future where everyone can fake everyone else. We're about to go through the looking glass in terms of trust in the information space and I don't think anyone understands the ramifications. - Read more in Import AI #97: https://t.co/FnF3vXldxd pic.twitter.com/nJqWrCXBuM
— Jack Clark (@jackclarkSF) June 6, 2018
Neat 📄 from #EuroVis:
— Mara Averick (@dataandme) June 6, 2018
"Exploring Interactive Linking Between Text and Visualization" ✒️ @beck_fabian &cohttps://t.co/tcvXcJvyCv #dataviz #infovis pic.twitter.com/71kWt6zdOh
Researchers develop AI that identifies and counts wildlife with 96.6% accuracy https://t.co/9JwjBQkHPz
— Nando de Freitas (@NandoDF) June 6, 2018
Look forward to reading this - have enjoyed his big picture think pieces (informed by deep experience in planning algorithms) in the past. "Model-free, Model-based, and General Intelligence," Hector Geffner: https://t.co/DyWjvwoJnH
— Miles Brundage (@Miles_Brundage) June 7, 2018
Tutorials and Resources
histbook: Versatile, high-performance histogram toolkit for Numpy. Cool stuff from the amazing Jim Pivarski of @diana_hep
— Kyle Cranmer (@KyleCranmer) June 6, 2018
Including nice plotting with Vega-Litehttps://t.co/xVkbTpaoDA
@vega_vis @CERN @CMSexperiment @ATLASexperiment @LHCbExperiment @jakevdp pic.twitter.com/i3HRrA2yDH
"encouraging psychologists to use mixed effects models is like giving shotguns to toddlers” - Altmann
— Indrajeet Patil (@patilindrajeets) June 6, 2018
In case you're worried about the validity of your model specification, use the R package `DHARMa` for (visual) residual diagnostics:https://t.co/H1QLP4RCPg#rstats #dataviz pic.twitter.com/OOJIwEBkcO
ICYMI, ✨ Harry Potter spells in R × 📱? There's an app for that…
— Mara Averick (@dataandme) June 6, 2018
"Making Magic w/ #Keras & Shiny" 🔮 @NicholasStrayer https://t.co/mIgDmTRh28 #rstats #rshiny pic.twitter.com/tcaX6Lywgm
Just released version 2.1 of Altair! Biggest new feature is easy specification of multi-value tooltips: https://t.co/Ku47PnmP0P
— Jake VanderPlas (@jakevdp) June 6, 2018
pip install -U altair pic.twitter.com/m3NRy7EpTg
Sentiment Use Across the Course of Pitchfork Music Reviews: A Tidy Text Analysis with R https://t.co/b2w74c6fLz #rstats #DataScience
— R-bloggers (@Rbloggers) June 7, 2018
Building a Deep Neural Network to play FIFA 18 https://t.co/SVnZV7scFS
— Nando de Freitas (@NandoDF) June 6, 2018
gganimate
gganimate have full support for ggraph (of course). Here I'm recreating an old temporal network animation using transition_events() pic.twitter.com/IelpUnNF1m
— Thomas Lin Pedersen (@thomasp85) June 6, 2018
case in point pic.twitter.com/bM0eG69qfW
— Thomas Lin Pedersen (@thomasp85) June 6, 2018
Code to produce a faceted versions: https://t.co/M9tgO479wP
— Thomas Lin Pedersen (@thomasp85) June 6, 2018
Miscellaneous
googled for an error message in pytorch, read my own answer from a year ago. full circle! Then dug around some stats. There are 34,300 posts on the PyTorch forums, viewed 7.6 million times. I wrote 1800 of them. So cool :D
— Soumith Chintala (@soumithchintala) June 7, 2018
Different interpretation of the same results: @jacobeisenstein HMM beats LSTM on small data @sleepinyourhat wow, LSTM beats HMM already with 500 sentences. Proof links: https://t.co/ao80M4NHhR https://t.co/wIxdqm5p93
— Leonid Boytsov (@srchvrs) June 6, 2018
Data scientists do three different jobs in varying combinations:
— Brandon Rohrer (@_brohrer_) June 5, 2018
Analysis – turning raw information into knowledge that can be acted on
Modeling – using the data we have to estimate the data we wish we had
Engineering – making everything else work faster, robustly, and at scale
Superintelligence
This essay is a must read for everyone thinking about AI predictions. https://t.co/lkzAmF8rBc
— Nando de Freitas (@NandoDF) June 7, 2018