The company will allow anyone to use certain tools that they implemented to make key improvements to the platform.
Facebook will allow anyone to download and make use of the same
Facebook will allow anyone to download and make use of the same artificial intelligence (AI) tools that the company used to make key improvements in the video and notification functions of the social network, as well as in its Messenger messaging application.
The software, which Facebook calls Horizon, will be available on GitHub, a public code storage platform, from now on, the company said in a blog post yesterday. GitHub is owned by Microsoft Corp.
Facebook used that set of tools internally to optimize the way that 360-degree videos are shown on the social network, taking into account factors such as the available bandwidth and how much of the video has already been loaded. According to the post on the blog, the same tools were used to better select what content to send to users through notifications. And they were used to refine the suggestions made by their intelligent assistant, called M, to the users of their Messengerapplication .
Horizon software focuses on reinforcement learning, by which it self-perfects by trial and error through experience to maximize some reward or minimize some loss and not through tagged data sets.
Reinforcement learning is the basis of a number of advances in AI, among which is the algorithm that defeated the best players in the world in the Go strategy game as well as those who now compete with humans in computer games. Multiple player complexes like Dota2 .
But until now, companies have rarely used it to solve real-world problems, in part because, outside games, it's often not prudent or safe to let an algorithm learn by trial and error. And for many real-world phenomena, there are no precise simulators in which an algorithm can be safely trained.
To elucidate what objective to give the algorithm, how to reward it for the actions that seem to lead to that goal and sanction it for actions that could have adverse consequences is also complicated outside of the games, where these elements are often incorporated into its structure.
To overcome these limitations, Facebook developed Horizon so that its teams could use reinforcement learning on the real problems facing the company, said Srinivas Narayanan , the company's director of applied machine learning, answering questions via email. But he added that the company now wanted to share the software with others.
"We are committed to open source, so it's a natural decision to share the latest system ready for production with the community," Narayanan said.
Facebook follows other AI research groups, including the AI DeepMindand GoogleBrain teams from Alphabet Inc. and OpenAI, which recently put reinforcing learning algorithms, programming tools and test environments at the public's disposal.
No comments:
Post a Comment