hey, we have 295 so far, and it is awesome, thank you!
It means that 15% of post viewers filled the questionnaire form of one question.

We kindly ask you to fill in if you haven’t yet, because we need 200 more responses to make this poll statistically significant!

💪🎆🎇 🎉we did it! 665 responses collected so far!
Thank you all. Results will be published as anonimized CSV (only countries and number of votes) later.

Jax is NumPy on CPU and GPU with automatic differentiation and JIT compilation.

Link: https://github.com/google/jax

And, channel statistics, as promised. You can look at the numbers at: https://docs.google.com/spreadsheets/d/1_y6ojxU7svUAmPViqWveUf0yVLqoZyukJTGikcC_5hQ.

And again, thank you for your support!

A Bag of Tricks for Image Classification

1. Large batch size
2. Mini model-tweaks
3. Refined Training Methods
4. Transfer Learning
5. Fancy Data Augmentation
https://link.medium.com/fzJvIBfsJS

#CV #tipsandtricks

Bayesian Optimization in AlphaGo

How latest AlphaGo agent win rate was improved from 50% to 66.5%.

ArXiV: https://arxiv.org/abs/1812.06855

​​Facebook have created and now open-sourced Nevergrad, a Python3 library that claims making easier to perform gradient-free optimizations.

Link: https://code.fb.com/ai-research/nevergrad/
Github: https://github.com/facebookresearch/nevergrad

Smart Compose: Using Neural Networks to Help Write Emails

Google shared some information about their new feature. Most important: they claim to focus on Fairness and Privacy, training on completely anonimized data and trying to eliminate biases.

Link: https://ai.googleblog.com/2018/05/smart-compose-using-neural-networks-to.html

#Google #SmartCompose #FairAI #Privacy

​​Reproducing high-quality singing voice
with state-of-the-art AI technology.

Some advance in singing voice synthesis. This opens path toward more interesting collaborations and sythetic celebrities projects.

P.S. Hatsune Miku's will still remain popular for their particular qualities, but now there is more room for competitors.

Link: https://www.techno-speech.com/news-20181214a-en

#SOTA #Voice #Synthesis

Deep learning cheatsheets, covering content of Stanford’s CS 230 class.

CNN: https://stanford.edu/~shervine/teaching/cs-230/cheatsheet-convolutional-neural-networks
RNN: https://stanford.edu/~shervine/teaching/cs-230/cheatsheet-recurrent-neural-networks
TipsAndTricks: https://stanford.edu/~shervine/teaching/cs-230/cheatsheet-deep-learning-tips-and-tricks

#cheatsheet #Stanford #dl #cnn #rnn #tipsntricks

​​Overview of current state of autonomously driving vehicle by Ben Evans.

Not so technical overview of where first autonomous vehicles will become commodity.

Link: https://www.ben-evans.com/benedictevans/2018/3/26/steps-to-autonomy

​​Creating super slow motion videos by predicting missing frames using a neural network, instead of simple interpolation. With code.

Github: https://github.com/avinashpaliwal/Super-SloMo
Website: https://people.cs.umass.edu/~hzjiang/projects/superslomo/

🎈🎊Geoff Hinton was named Companion of the Order of Canada (highest Canadian honor).

Hinton — one of the pioneers of the Deep Leaning, author of the famous «Neural Networks» on Coursera, one of the creators of the modern ML industry.

Congratulations!

https://www.gg.ca/en/media/news/2018/governor-general-announces-103-new-appointments-order-canada

​​Scaling Uber’s Apache Hadoop Distributed File System for Growth

Post on how #Uber team handles #Hadoop challenges.

https://eng.uber.com/scaling-hdfs/

#BigData #HDFS

​​Massively Multilingual Sentence Embeddings for Zero-Shot Cross-Lingual Transfer and Beyond

New SOTA on cross-lingual transfer (XNLI, MLDoc) and bitext mining (BUCC) using a shared encoder for 93 languages.

Link: https://arxiv.org/abs/1812.10464

#SOTA #NLP

​​Building Automated Feature Rollouts on Robust Regression Analysis

Nice article on important thing — statistical analysis of hypothesis testing. Every new feature or change made to existent one is basically an experiment. Article covers how #Uber team handles this in live system.

Link: https://eng.uber.com/autonomous-rollouts-regression-analysis/

#Uber #statistics #production #truestory

Battling Entropy: Making Order of the Chaos in Our Lives

Article on #entropy as a concept.

Link: https://fs.blog/2018/11/entropy/

A disciplined approach to neural network hyper-parameters

Recommendations on how to optimize learning rate, weight decay, momentum and batch size.

ArXiV: https://arxiv.org/pdf/1803.09820.pdf

#nn #hyperopt