Day 49: SwitchNorm

Funding probabilities on Shark Tank, grouped by gender. #datavizhttps://t.co/GYwvyYEE5I pic.twitter.com/yYkv5dHiaU

— Randy Olson (@randal_olson) July 7, 2018

Estimated Steam game player counts (Source: Steam Spy) pic.twitter.com/4Oztsw1uhZ

— Mike Bostock (@mbostock) July 7, 2018

What does it mean to say "there's a real possibility that football's coming home"? Lovely visualisation:https://t.co/WYYuTMx3mj

— Tim Harford (@TimHarford) July 7, 2018

The quest for optimal normalization in neural nets continues. SwitchNorm: add BatchNorm + InstanceNorm + GroupNorm with a learnable blend at each layer https://t.co/9hOQXnkk8T fun plots; + code https://t.co/34r96BStCS

— Andrej Karpathy (@karpathy) July 7, 2018

Interesting! Adding that to the reading list. Is there a particular reason they didn't include the recent GroupNorm by Yuxin Wu & Kaiming He? https://t.co/UMSiG3kxXU

— Sebastian Raschka (@rasbt) July 8, 2018

Will present my #midl2018 poster next session. Learn how to train a normal CNN with 8192x8192 input sizes and a single label on one GPU! (from 235gb to 7gb memory required)
Code: https://t.co/KzDPFnbArl pic.twitter.com/Vcfco5fBPk

— Hans Pinckaers (@hanspinckaers) July 6, 2018

How to Backdoor Federated Learning, Bagdasaryan et al. – attacks federated learning scenarios where many users contribute to a single shared model: https://t.co/7liNzrenBa pic.twitter.com/c8f7xEds04

— Brendan Dolan-Gavitt (@moyix) July 7, 2018

Qualitative Data Science: Using RQDA to analyse interviews https://t.co/2hMIcscAMD #rstats #DataScience

— R-bloggers (@Rbloggers) July 7, 2018

The official repository for the Deep Reinforcement Learning Nanodegree program at @udacity is now public! Check it out to see many implementations in @PyTorch, including DQN, DDPG, and much more! https://t.co/umOUxLdwTj pic.twitter.com/Be5UAILQ6l

— Alexis Cook (@alexis_b_cook) July 6, 2018

How many random seeds are needed to compare #DeepRL algorithms?

Our new tutorial to address this key issue of #reproducibility in #reinforcementlearning

PDF: https://t.co/7eHOzhtLuC

Code: https://t.co/0CRRM8RYYr

Blog: https://t.co/rYWM5zPYZB#machinelearning #neuralnetworks

— Pierre-Yves Oudeyer (@pyoudeyer) July 6, 2018

ICYMI, 👩‍🏫 great material – code, slides, & 🎬!
💻 "Code for Workshop: Intro to Machine Learning w/ R" by @ShirinGlander https://t.co/An6MvGx4TH #rstats #MachineLearning pic.twitter.com/tpigxcH4vT

— Mara Averick (@dataandme) July 7, 2018

"Pandas on Ray – Early Lessons from Parallelizing Pandas" - almost forgot about this neat project! https://t.co/3CKXWlPHo7

— Sebastian Raschka (@rasbt) July 8, 2018

Horovod – distributed training framework for TensorFlow, Keras, and PyTorch
By @UberEng

Require far less code changes than the Distributed TensorFlowhttps://t.co/ljrWzTJZ4b #MachineLeaning pic.twitter.com/gbLTRUVtWc

— ML Review (@ml_review) July 8, 2018

This tutorial will show you how to convert a neural style transfer model that has been exported from @PyTorch and into the #CoreML format using ONNX. #AI #MachineLearning #Developers https://t.co/AXVuTolAbP

— ONNX (@onnxai) July 7, 2018

How did I only just find out about this package??? https://t.co/aXJe2zLRC3. You have all let me down!

— Hadley Wickham (@hadleywickham) July 6, 2018

probably shouldn't share it publicly, but I have too many projects to work on at the moment anyway: had a great idea to improve ELMs further, ie dropping the hidden layer(s) + run n ELMs with n different random seeds to construct a majority vote ensemble. Someone should try this!

— Sebastian Raschka (@rasbt) July 8, 2018

A useful paradox for data scientists to keep in mind:

Most cities are small, but most people live in large cities

This relates to analyses of e.g. user engagement: most of your users probably don't do much, but most of your engagement is from frequent users

— David Robinson (@drob) July 5, 2018

An astonishing paper that may explain why it’s so difficult to patch.

They monitored 400 libraries. In 116 days, they saw 282 breaking changes!

Each day, there’s 6.1% chance of breaking chg, for each lib you use!@topopal @mtnygard @mik_kersten @ctxthttps://t.co/qdRswgAwm2 pic.twitter.com/sns2IOoK0L

— Gene Kim (@RealGeneKim) July 5, 2018