Last Week in AI

Source: Deep Learning on Medium


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: Towards a Private AI

The biggest dilemma of the next decade of artificial intelligence(AI) is to determine whether data will remain controlled by large companies or will be democratized across its real owners. AI models regularly use data produced by users that can’t benefit from this asset or even controlled access to it. Privacy models are necessary for AI to achieve mainstream adoption.

In recent years, there have been an resurgence in the interest for privacy-preserving AI models. Recently, OpenMined released a new privacy framework called PySyft that enables privacy computations in popular deep learning stacks such as PyTorch, TensorFlow or Keras. This week, our team at Invector Labs published a comprehensive analysis of PySyft as well as the principles behind private AI.

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

AI Research

Microsoft AI researchers published two papers discussing ideas for improving the reliability of reinforcement learning models. Reinforcement learning is the technique behind great AI successes such as AlphaGo or self-driving vehicles.

>Read more in this blog post from Microsoft Research

The Uber data science team published a fascinating post about how they use data to improve the interaction with pedestrians in self-driving vehicles.

>Read more in this blog post form the Uber engineering team

Researchers from the University of Stuttgart published a research paper to address the challenge of forgetting when training neural networks in sequential tasks.

>Read the research paper here

Cool AI Tech Releases

Google released a reinforcement learning environment for teaching AI agents how to play soccer/football. The environment allows you to configure an AI agent to learn to play football simply by playing.

>Read more in this blog post from Google research

Staying with Google, the internet giant also released TensorNetwork, a new open source library for quantum calculations. Quantum scenarios such as energy calculations are becoming more important in machine learning applications.

>Read more in this blog post from Google research

Microsoft released a new natural language processing library for phonetic matching. The library was previously created by Maluuba, a NLP startup Microsoft acquired two years ago.

>Read more in this blog post from Microsoft Research

AI in the Real World

Amazon hosted its first re:MARS conference this week showcasing advances in robotics, AI and space travel.

>Read more in this coverage from The Seattle Times

Buzzfeed News has linked U.S.-based organizations like the Alaska Retirement Management Board, Rockefeller Foundation, and Princeton University to private equity and venture capital firms funding some of China’s AI startups used for state surveillance.

>Learn more in this article from Buzzfeed

AI researchers created a version of Reddit in which the participants are bot simulations of popular Reddit handles.

>Read more in this fascinating article from The Verge