Engineers have trained LASER, a language model that embeds data from 93 languages as vectors and maps them onto the same latent space, to help translate between rarer language pairs. The hope is that Facebook will be better at detecting hate speech and online bullying across multiple languages.
For visual content, Facebook uses a mixture of things. It has a computer vision model called a panoptic feature pyramid network to segment images by objects so it can identify things in the foreground and background. In videos, Facebook has trained software to understand specific actions from 65 million public Instagram videos to identify violent or graphic footage.
They may sound impressive, but it’s trickier dealing with real content that deviates from the training data. For example, Facebook failed to take down the video of the Christchurch mosque shootings being live streamed.
Read more
Original article written by Katyanna Quatch
For more articles like this, visit us at:
https://www.victoriousds.com
No comments:
Post a Comment