Facebook trains computers to learn like humans

Facebook trains computers to learn like humans


The company's chief AI scientist Yann LeCun has opened up on the company's goals to make machines learn through observation in the same way babies do.

As quoted by CNET, he said:

"We'd like artificial intelligence systems to learn how the world works by observation because that will have a huge implication. It would allow machines to have some level of common sense."

The social media brand's team has pushed computers towards the goal of filling in the blanks themselves without needing humans to sort through data. This could help Facebook itself by improving the likes of content moderation.

On Thursday (04.03.21), the AI team revealed it had achieved a "breakthrough" with its self-supervised computer vision model Seer. Seer - short for SElf-superERvised - learned from a billion random, uncurated and unlabeled Instagram images.

The model was able to correctly identify and categorise the dominant object with an 84.2% accuracy rate. According to the group's study, Seer outperformed the top existing systems by 1%.

In a blog post, they explained:

"[These finds are a] major breakthrough that ultimately clears the path for more flexible, accurate and adaptable computer vision models in the future."


And Yann added:

"The advantage of self-supervised learning is that you can train very big networks and it will still be accurate."

Copyright © 2009-2021 BANG Media International

- Advertisement -
Dark Light