Every week, my team at Invector Labs publishes a newsletter to track the most recent developments in AI research and technology. You can find this week’s issue below. You can sign up for it below. Please do so, our guys worked really hard on this:
From the Editor
Building machine learning systems at scale is an incredibly daunting task. For all the advancements in machine learning research and technology, the stacks and best practices for developing large scale systems remain a very nascent state. While data scientists can rapidly pilot new algorithms using frameworks like TensorFlow or PyTorch, the process of deploying and scaling those models is a very difficult challenge.
Among the large internet companies pushing the boundaries of machine learning, Uber has been incredibly public about their architecture and best practices. In just the last year, the Uber engineering team has unveiled technologies like Michelangelo, Pyro.ai and Horovod which all focus on solving critical areas of the lifecycle of machine learning applications in the real world. This week, Uber unveiled PyML, a new library for rapidly implementing machine learning models in a way that can be production-ready. Certainly, the lessons from Uber can be incredibly valuable for organization starting their machine learning journey.
Now let’s take a look at the core developments in AI research and technology this week:
Google proposes a method to better annotate images to train deep learning models
OpenAI published a new paper proposing a method for modeling complicated behaviors that exceed human scale by decomposing them into simpler tasks.
IBM publishes a new method for training AI agents on behaviors that align with “societal values”
Cool Tech Releases
Uber engineering unveiled PyML, an extension to its Michelangelo platform that enables the rapid development of machine learning models
AI in the Real World
The robot and the fragrance: IBM has partnered with Symrise to create new machine learning models that improve the creation of perfumes.
Using 12 neurons to park a car: AI researcher at the Tu Wien institute developed a surprisingly simple method to park a vehicle.
Fixing transportation in NYC: AI students participating in Microsoft’s data science summer school took a fresh look at how to use AI to improve NYC transit system.