Data Science by ODS.ai 🦜
51K subscribers
363 photos
34 videos
7 files
1.52K links
First Telegram Data Science channel. Covering all technical and popular staff about anything related to Data Science: AI, Big Data, Machine Learning, Statistics, general Math and the applications of former. To reach editors contact: @haarrp
Download Telegram
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
Plug-and-play differential privacy for your tensorflow code

#GoogleAI has just released a new library for training machine learning models with (differential) privacy for training data.

where you would write tf.train.GradientDescentOptimizer
instead just swap in the DPGradientDescentOptimizer


Tutorial: https://github.com/tensorflow/privacy/blob/master/tutorials/mnist_dpsgd_tutorial.py
Link: https://github.com/tensorflow/privacy

#Privacy #tensorflow
​​TOWARDS FEDERATED LEARNING AT SCALE: SYSTEM DESIGN

Google published how they do #FederatedLearning at scale on tens of millions of mobile phones. This is about training model on decentralized data.

ArXiV: https://arxiv.org/pdf/1902.01046.pdf

#Google #Privacy
Estimating the success of re-identifications in incomplete datasets using generative models

99.98% of Americans would be correctly re-identified in any dataset using 15 demographic attributes, suggesting that even heavily sampled anonymized datasets are unlikely to satisfy the modern standards for anonymization set forth by GDPR.

This is a big concern about privacy and a problem for Data Engineering, especially for those working with anonymized personal information. Paper provides a way to re-identify person from anonymized dataset, this can be useful for people who work for government or security companies

https://www.reddit.com/r/science/comments/chko43/9998_of_americans_would_be_correctly_reidentified/

#privacy #gdpr #federatedlearning #ml