DeepMind is developing one algorithm to rule them all

That Transform Technology Summits launch on October 13 with Low-Code / No Code: Enabling Enterprise Agility. Register now!


DeepMind wants to enable neural networks to mimic algorithms to get the best of both worlds, and it uses Google Maps as a test bed.

Classical algorithms are what have enabled software to eat the world, but the data they work with does not always reflect the real world. Deep learning is what drives some of the most iconic AI applications today, but deep learning models need to be retrained to be applied to domains they were not originally designed for.

DeepMind tries to combine deep learning and algorithms and creates the one algorithm to control them all: a deep learning model that can learn to emulate any algorithm and generate an algorithm-equivalent model that can work with real data.

DeepMind has made headlines for some iconic AI achievements. After developing AlphaGo, a program that became world champion at the game Go in a five-match battle after beating a human professional Go player, and AlphaFold, a solution to a 50-year major challenge in biology, DeepMind has set its sights on another major challenge: bridging deep learning, an AI technique with classical computer science.

Birth of neural algorithmic reasoning

Charles Blundell and Petar Veličković both hold senior research positions at DeepMind. They share a background in classical computer science and a passion for applied innovation. When Veličković met Blundell at DeepMind, a line of research known as Neural Algorithmic Reasoning (NAR) was born after the uniform position paper recently published by the duo.

The central thesis is that algorithms possess fundamentally different qualities for deep learning methods – something Blundell and Veličković elaborated on in their introduction of NAR. This suggests that if deep learning methods were better able to mimic algorithms, generalization of the kind seen with algorithms would be possible with deep learning.

Like all well-founded research, NAR has a pedigree that goes back to the roots of the fields it touches on and goes out to collaborate with other researchers. Unlike much pie-in-the-sky research, NAR has some early results and applications to show.

We recently sat down to discuss the first principles and foundations of NAR with Veličković and Blundell, who will also be joined by MILA researcher Andreea Deac, who expanded on details, applications and future directions. Areas of interest include processing graphical data and pathfinding.

Pathfinding: There is an algorithm for it

Deac interned at DeepMind and became interested in learning graph representation through the lens of drug discovery. Learning graph representation is an area Veličković is a leading expert in, and he believes it is a great tool for processing graph data.

“If you squint hard enough, any kind of data can fit into a graph representation. Images can be seen as graphs of pixels associated with proximity. Text can be seen as a series of objects linked together. More generally, things are really coming from nature, which is not designed to fit within a frame or in a sequence that humans would do, in fact quite naturally represented as graph structures, ”said Veličković.

Another real problem that lends itself well to graphs — and a standard problem for DeepMind, which, like Google, is part of Alphabet — is road search. By 2020, Google Maps was the most downloaded map and navigation app in the United States and used by millions of people every day. One of its killer features, Pathfinder, is powered by none other than DeepMind.

The popular app now shows an approach that can revolutionize AI and software as the world knows them. Google Maps has a real-world road network that helps predict travel times. Veličković noted that DeepMind has also been working on a Google Maps application that uses graphing networks to predict travel times. This now shows queries in Google Maps worldwide, and the details are described in a recent publication.

VentureBeat

VentureBeat’s mission is to be a digital urban space for technical decision makers to gain knowledge about transformative technology and transactions. Our site provides important information about data technologies and strategies to guide you as you lead your organizations. We invite you to join our community to access:

  • updated information on topics that interest you
  • our newsletters
  • gated thought-leader content and discount access to our valued events, such as Transform 2021: Learn more
  • networking features and more

sign up

Leave a Comment