Last Week in AI

Source: Deep Learning on Medium

Every week, my team at Invector Labs publishes a newsletter that covers 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: What AlphaStar Means for AI

Earlier this year, artificial intelligence(AI) powerhouse DeepMind unveiled an earlier version of AlphaStar, a reinforcement learning agent that could challenge the top world players in StarCraft II. This week, DeepMind published an update showing that AlphaStar can now play the full game at a Grandmaster level under professionally approved conditions. Specifically, AlphaStar was able to raise to at Grandmaster level in all StarCraft modalities overperforming 99.8% of human players.

Even though AlphaStar is focused on a gaming scenario, the development can have profound implication for general-purpose AI solutions. StarCraft is a strategy game in which agents need to collaborate, cooperate, explore new environment and make decisions in the absence of complete information. From self-driving vehicles to economic planning, the principles behind AlphaStar can become foundational to a new generation of AI applications.

Now let’s take a look at the core developments in AI research and technology this week:

AI Research

Playing StarCraft Without Any Restrictions

DeepMind published a new research describing AlphaStar II, the first AI that plays StarCraft at a Grandmaster level.

>Read more in this blog post from DeepMind

Transferring Knowledge Across Languages

Amazon researchers published a paper proposing a method to reduce the data required for transfer learning in natural language understanding models.

>Read more in this blog post from Amazon Research

Learning to Assemble and Disassemble Objects

Google published a research paper outlining a new method for training a robot to disassemble and assemble objects.

>Read more in this blog post from Google Research

Cool AI Tech Releases

A Dataset to Train Conversational Assistants

Google released a dataset for training AI agents on task-oriented conversational dialogs.

>Read more in this blog post from Google Research

TensorFlow Gets Collaborative

Google launched TensorBoard.dev, a new environment for sharing machine learning experiments.

>Read more in this coverage from VentureBeat

Parallel Neural Network Training

Microsoft Research open source PipeDream, a new framework for scaling the training of deep learning models.

>Read more in this blog post from Microsoft Research

AI in the Real World

AI and the Future of Work

IBM and MIT published a study about how AI is impacting specific jobs.

>Read more in this blog post from IBM Research

Solving a Famous Math Problem 100 Million Times Faster

AI researchers form the University of Edinburgh build a series of neural networks that solved the famous three-body problem 100 million times faster than previous attempts.

>Read more in this coverage from MIT Technology Review

US DOE Gets Serious About AI

The US Department of Energy is pushing a multi-billion AI initiative to improve scientific discoveries.

>Read more in this coverage from Science Magazine