Bits & Bytes

A Facebook research program into multi-agent negotiations resulted in bots creating their own streamlined ways to communicate. The truth behind the ballyhoo about those language-inventing Facebook bots.
Meanwhile, Google DeepMind researchers are working on developing agents that imagine the future and plan their tasks in two new papers focused on “imagination-based planning.”
If all this freaks you out, researchers from Cornell, Univ. of Montreal and Univ. of Louisville published “Guidelines for Artificial Intelligence Containment,” in which they propose guidelines for helping AI researchers develop reliable sandboxing software for intelligent programs.
In January of this year, Google Cloud hired Fei-Fei Li, director of the Stanford AI & Vision labs, as its chief scientist for AI & ML. Dr. Li is perhaps best known for her role in creating the ImageNet dataset upon which much of the recent progress in object recognition has been based. A couple of months later, in March, at their Next conference, Google Cloud announced its acquisition of Kaggle. With this context, it’s not too surprising that Kaggle just announced that it’ll be hosting all 3 ImageNet challenges for the first time ever.
Fast.ai announced the availability of Part 2 of their highly regarded (and FREE) online course, Deep Learning for Coders. Topics include TensorFlow and style transfer, generative models and GANs, memory networks, attentional models and more.
Things continue to heat up in the AI startup ecosystem: Google announced Launchpad Studio, a new accelerator program for AI & ML startups, while autonomous vehicle startup Momenta closed a $46M series B financing, and AI-for-robots company Vicarious closed a $50M series C.