There are so many great tweets today that it’s almost overwhelming. Enjoy!
Awesome Collection of Paper Summaries
(All written within one month! Very impressive.)
In May, I decided to read & summarize a #machinelearning paper every day. I'll curate some of these into blog posts, but in the meantime, I'd like to share the full document, where I delve into GAN tricks, innovative word embeddings, LSTM structure, & more https://t.co/rX0xnt3mlY
— Cody Wild (@decodyng) June 1, 2018
AI Tulips
More AI tulips!!! With the larger dataset I’ve made and spectral normalisation I’m able to get much more variety in the type of tulips it produces. pic.twitter.com/19vhpyLwDj
— anna ridler (@annaridler) June 1, 2018
And here it is training from epochs 1 -250 pic.twitter.com/Al39E0zymR
— anna ridler (@annaridler) June 1, 2018
From Tulip Bubble to AI Bubble 🌷 https://t.co/wdnPMdL0tn
— hardmaru (@hardmaru) June 1, 2018
#CraftyDataViz
Announcing the winners of the #CraftyDataViz contest!
— Data Science Renee (@BecomingDataSci) June 2, 2018
CONGRATULATIONS to @pattithepotato, @AmyCesal, and @jschwabish's daughter!https://t.co/yL1Lbhj68s pic.twitter.com/TAxUsazQvs
THANK YOU to everyone who entered - the entries were ✨AMAZING✨ and so hard to choose from! And thank you again to the judges that tackled that difficult task:https://t.co/kBoUDkN8XD
— Data Science Renee (@BecomingDataSci) June 2, 2018
User Experience Design for APIs
Every time I start a new project or design a new API. First thing I do is read this incredibly important Design Principles. @fchollet Thank you for this and Keras! https://t.co/Vjna3DwhlW
— SandeepKrishnamurthy (@skm4ml) June 1, 2018
Tensorly
(Tensor Learning in Python)
Random matrix decompositions are amazingly great - especially now you can do them on the GPU with https://t.co/DEiovlrjgF. Check out @JeanKossaifi's timing comparison below, and read Halko's wonderful paper for the foundational work: https://t.co/VS3Us1N43j https://t.co/JZ4LlntLhz
— Jeremy Howard (@jeremyphoward) June 1, 2018
Factorising a 1 billion element low-rank tensor (1000x1000x1000), revisited.
— Jean Kossaifi (@JeanKossaifi) June 1, 2018
If you are short on compute and/or memory, try the randomised version!
Ran with #TensorLy, #numpy and @PyTorch on @awscloud Tesla V100. pic.twitter.com/THWZpwA4hg
[Ethics] On Gender Imbalance in Conferences
Men, if you're ever on the organizing committee for a conference, and the committee is made up of all men, suggest that they invite some ladies to participate in organization. And if they say they don't want to add anyone, step down "to make room" for them to add women.
— Data Science Renee (@BecomingDataSci) June 2, 2018
Do you also advocate that female professors of veterinary medicine quit their jobs and give them to men? https://t.co/7cKq0LRplT https://t.co/2fXrq9vonQ
— Amy Alkon (@amyalkon) June 2, 2018
Oh hi.
— Data Science Renee (@BecomingDataSci) June 2, 2018
(Hint: the "to make room for" was in quotes, because I think they should quit because of not wanting to sit on a conference organizing committee - which is typically volunteer work - that is not inclusive)
Answer to your question: No.
The assumption that they've literally asked all qualified women is laughable & it also shows how limited their personal networks are, which is why having more diverse participation is helpful for that aspect.
— Data Science Renee (@BecomingDataSci) June 2, 2018
This is a bunch of BS, and I'm not falling for this line of reasoning.
Are you looking to attend an all-male conference in #DataScience? #rfinance2018 has got you covered! 🧑🏻🙋🏻♂️🧓🏻👱🏻♂️🤵🏻👨🏻🧔🏻👴🏻👨🏻💼
— Women in ML/DS (@wimlds) June 1, 2018
100% male committee, 100% male speakers, no Code of Conduct. Yes, this is 2018! 📆 https://t.co/EfhR1QhwWj #BinderFullofMen 👬 pic.twitter.com/NLbS31y43V
Diversity and inclusion *require* intentionality… Great write-up by @Lady_Ada_King https://t.co/YFMPxfxETn
— Mara Averick (@dataandme) June 1, 2018
[Ethics] Google Pulls Out of Project Maven
When Google pulls out of Project Maven, who is mostly likely to fill the void? My gut feeling is Amazon, since there seems to be little Amazon formally has qualms about, but maybe there's someone else interested in the gig.
— Kelsey D. Atherton (@AthertonKD) June 1, 2018
This is why, if Google wants to be serious about ethics + AI, it should be involved in discussions of governing lethal autonomous weapons, etc. Not participating is insufficient if the goal is to make something not happen. https://t.co/3hyqGkAiXs
— Miles Brundage (@Miles_Brundage) June 1, 2018
I think Google’s voice could be quite influential in such debates and have a much broader impact. Alas, besides this leaked internal debate, the only loud voice they’re really giving off externally is Eric Schmidt’s hawkishness. Much more to be done.
— Miles Brundage (@Miles_Brundage) June 1, 2018
(from https://t.co/JCbduYt5eb) what does that even mean? Better ML for drone footage processing = 100x safer US troops? Umm
— Miles Brundage (@Miles_Brundage) June 1, 2018
Notable Research
Do Better ImageNet Models Transfer Better? https://t.co/tSZS25HK2p
— Denny Britz (@dennybritz) June 1, 2018
It’s not the goal of the paper, but for us people who don’t follow the CV SOTA race this is a neat cheat of model performance. pic.twitter.com/4wJCSBX3Lp
Our recent work on the role of over-parametrization in generalization of neural nets:https://t.co/7UguNspEeo
— Behnam Neyshabur (@bneyshabur) May 31, 2018
This helps us to understand the phenomenon we reported more than 3 years ago (also observed by some people before deep learning era):https://t.co/zPAxAP2Q7y pic.twitter.com/LsqqeBCPxV
Towards de novo drug discovery using a molecular graph generative model combined with chemical metrics optimization (WGAN + RL). Check out our new MolGAN: an implicit generative model for small molecular graphs (https://t.co/kDh8NpwQ5L ) with @thomaskipf pic.twitter.com/KOTJ6jngE2
— Nicola De Cao (@nicola_decao) May 31, 2018
Good morning! I'm very happy to say that our (Kang Liu, me, and @sg175) paper on automatically removing backdoors from DNNs has been accepted to RAID 2018! You can find a pre-print here: https://t.co/FQKaGT3MGM Here's a quick thread on how it works... pic.twitter.com/XoxgQEsCo1
— Brendan Dolan-Gavitt (@moyix) June 1, 2018
Reward design in RL is a bit of a mystical art. We can start to demystify it using control as inference, and also learn rewards from data using variational inverse control with events (VICE): https://t.co/VmUeMLg4PN
— Sergey Levine (@svlevine) June 1, 2018
w/ Justin Fu, @avisingh599 Dibya Ghosh, Larry Yang
Our recent paper with Vineet Gupta and @philipmlong https://t.co/DB1MUv2O5i
— Hanie Sedghi (@HanieSedghi) June 1, 2018
This is very thoughtful work on cooperative inverse reinforcement learning https://t.co/B3uCwr1Qfc
— Nando de Freitas (@NandoDF) June 1, 2018
Using Artificial Intelligence to Develop Electricity Load Forecasts
— Iain Brown, PhD (@IainLJBrown) June 1, 2018
Read more here: https://t.co/ymLHnbda4l#ArtificialIntelligence #AI #DataScience #MachineLearning #BigData #DeepLearning #NLP #Robots #IoT
"To Trust Or Not To Trust A Classifier" by Google Research https://t.co/mgvaXTTiea : beyond simple confidence scores. The ability to auto-detect bad predictions in critical for safe deployments in sensitive applications. #MachineLearning #DataScience #AI
— Nenad Tomasev (@weballergy) May 31, 2018
Tutorials / Resources
Complete draft of a new textbook for NLP: https://t.co/UNnl5n5oG4
— Jacob Eisenstein (@jacobeisenstein) June 1, 2018
Thanks to everyone who gave me edits and corrections! Stop me at #NAACL2018 and I'll buy you a beer or a beignet.
Here are the data and code for that study of Puerto Rico deaths https://t.co/pLgvcIivpf
— Andrew Gelman (@StatModeling) June 1, 2018
For all the folks who are preparing for data science interviews like me, here is a notebook on the implementation of common data structures and algorithms that might be useful.
— Shikhar Gupta (@shik1470) June 1, 2018
NBviewer: https://t.co/QBGzxfPQTI
Github: https://t.co/85t1jNV7Js
Building a Question-Answering System from Scratch using Facebook Research's Sentence Embeddings, by @alvira_swalin @usfca_msds https://t.co/8qKYI9SKMD
— Rachel Thomas (@math_rachel) June 1, 2018
I went through the Horovod codebase -- I really like it. Clever, clean, simple. Simplicity is the highest quality of software.https://t.co/RByCXitVg4
— François Chollet (@fchollet) June 1, 2018
ICYMI, 😻 the handiest:
— Mara Averick (@dataandme) June 1, 2018
"Probability Cheat Sheets" by @wzchen & @stat110 https://t.co/q7aIsGIk1f #probability #statistics #SoDS18 pic.twitter.com/XiSStJuxhd
In May, I decided to read & summarize a #machinelearning paper every day. I'll curate some of these into blog posts, but in the meantime, I'd like to share the full document, where I delve into GAN tricks, innovative word embeddings, LSTM structure, & more https://t.co/rX0xnt3mlY
— Cody Wild (@decodyng) June 1, 2018
This week's #KernelAwards winner uses linear regression to predict happiness scores based off of independent variables such as family, economy, life expectancy, and freedom: https://t.co/Y1XMAFct3L pic.twitter.com/GYEVdKpytp
— Kaggle (@kaggle) June 2, 2018
Nice (large) driving dataset from Berkeley. https://t.co/d6isNnaA8T
— Yann LeCun (@ylecun) June 1, 2018
#rstats
🎊 Just a friendly reminder that @sharon000 has *super* helpful #rstats guides… https://t.co/NffvvKV0Pm #r4ds pic.twitter.com/vGWy2AMLHI
— Mara Averick (@dataandme) June 1, 2018
June is #ShinyAppreciation Month!#Shiny is an R-package that makes it easy to build interactive web apps straight from R. Lesson 1 of 7, in this lesson will get you started building Shiny apps right away. https://t.co/0owbTFLShq pic.twitter.com/VUu2IKQxVi
— RStudio (@rstudio) June 1, 2018
Welcome #ShinyAppreciation month! Here is an open #Shiny app that allows to explore tweets by all members of the European Parliament, create word clouds, and customise outputs. Check it out https://t.co/BlbxkhIcaj @EdjNet pic.twitter.com/dL74fnAQWW
— Giorgio Comai (@giocomai) June 1, 2018
Miscellaneous
A better motivational poster: "Move fast and make sure all CI checks are green before merging"
— François Chollet (@fchollet) June 2, 2018
You can learn more about PAIR at https://t.co/ESGtQSwfjd
— Jeff Dean (@JeffDean) June 1, 2018
The most interesting thing about the "Thanksgiving Effect" study is what it tells us about the limits of data anonymizationhttps://t.co/Iy8YEigZfV pic.twitter.com/fxT8p39rzH
— Cory Doctorow (@doctorow) June 1, 2018
"Even without high-tech surveillance, Xinjiang’s police state is formidable. With it, it becomes terrifying.”
— Julia Angwin (@JuliaAngwin) June 1, 2018
Important reminder of how surveillance is so often used to oppress minority groups.https://t.co/thAIjTQfOG
I’m 😍😍 over Danny Kaplan’s historic lesson on stats. 1908— the t-test and the Model T, 1920s, ANOVA and the Model A. pic.twitter.com/jlUwsVpZpr
— Amelia McNamara (@AmeliaMN) June 1, 2018
So, Facebook, I'm very suddenly seeing adverts for boutique Amsterdam hotels because I speak English (US) and am over 26? Not because I just booked a hotel for Amsterdam next week on a different site, even with a tracker blocker installed? pic.twitter.com/uVkOUkWXJJ
— Michael Veale (@mikarv) May 31, 2018
As I've noted before, when reading Nick Bostrom's book "Superintelligence", you can replace "superintelligence" with "unregulated capitalism" and it makes perfect sense.
— Thomas G. Dietterich (@tdietterich) May 31, 2018
Check out my new @TheAtlantic article using sociological research to explain why the classic marshmallow test didn't hold up under closer scrutiny.
— Jess Calarco (@JessicaCalarco) June 1, 2018
And thanks to @tw_watts for the new study that inspired this piece! https://t.co/elttBCz6DC