True Positive Weekly #9

The most important artificial intelligence and machine learning links of the week

In this issue: graph neural networks help Google improve the accuracy of ETA on Google Maps by 50%, Google's BigBird is a Transfomer for even longer sequences, 50 new planets confirmed in machine learning first, scraping court records data to find dirty cops, the dumb reason your AI project will fail, and more.

• DeepMind uses graph neural networks to boost Google Maps ETA accuracy by up to 50% (official blog)

• The Incredible PyTorch: a curated list of tutorials, papers, projects, communities, and more relating to PyTorch

• Google's BigBird allows Transformer neural networks to process sequences up to 8x longer than previously possible

• 50 new planets confirmed in machine learning first

• Reproducing 150 research papers in machine learning: the problems and solutions

• Familiar faces but a different voice: a Wav2Lip demo

• AI researchers use heartbeat detection to identify deepfake videos

• Scraping court records data to find dirty cops

• The dumb reason your AI project will fail

• [Announcement] The Machine Learning Engineering book by Andriy Burkov is now released on Leanpub and Amazon

Enjoy the newsletter? Help us make it bigger and better by sharing with your colleagues and friends.