DeepMind develops a universal architecture for building artificial intelligence

Researchers at DeepMind have built the universal Perceiver IO architecture to handle all types of input and output data.

To tackle all the challenges we meet while solving intelligence, we need tools that are as adaptable as possible. Announcing the paper & code for Perceiver IO, an architecture that handles a wide range of data and tasks, all while scaling gracefully: https://t.co/9UU9b6276q 1/4 pic.twitter.com/PO9JSHt7yL

— DeepMind (@DeepMind) August 3, 2021

At its core is the original Perceiver model introduced in June 2021. It handles images, audio, video and their combinations, but is limited to tasks with simple outputs, such as classification.

To address this limitation, researchers created a more general version of the architecture — Perceiver IO. It can produce a wide range of outputs from diverse input data, making it applicable to fields such as natural language processing, computer vision and multimodal understanding.

Perceiver and Perceiver IO are built on transformer architectures that work well for input data containing thousands of elements. However, according to the researchers, images, audio and video can contain millions of such elements.

“With the original Perceiver, we solved the main problem of a universal architecture: scaling transformers to very large input data without introducing domain-specific assumptions,” the blog states.

Researchers also believe that Perceiver IO could reach an unprecedented level of universality.

They published the architecture’s source code on GitHub and hope it will help researchers and practitioners develop applications without having to spend resources on bespoke solutions built with specialized systems.

Recall that in late July DeepMind introduced the extensive gaming environment XLand for training universal artificial intelligence agents.

In July, the AI lab team collected and published the most complete database of human protein structures, created by the neural network AlphaFold.

In June, scientists from DeepMind stated that reinforcement learning is sufficient to achieve artificial general intelligence.

Subscribe to ForkLog news on Telegram: ForkLog AI — all the news from the world of AI!