Thanks to progress in its artificial intelligence research, Facebook FB is on track to make news feeds even more attuned to users’ tastes.
The social network said on Tuesday that it has created a system that teaches machines to identify objects and actions in a photo that is 30% faster and uses 10 times less training data than most other systems and industry benchmarks. Facebook’s system also has a new feature that allows people to ask the machine what is in a photo, which the company calls “VQA” or visual Q&A. This feature is supported by Facebook’s natural language understanding technology called “Memory Networks,” which infers answers to simple questions after reading text. The technology could be used to improve Facebook’s ability to serve up images, posts and ads that resonate best with users and will likely also be valuable to visually impaired users. Facebook has also been using artificial intelligence to help power its new personal assistant tool, “M,” housed in Messenger.
“Our artificial intelligence (AI) research efforts — along with our work to develop radical new approaches to connectivity and our work to develop immersive new VR technologies — are a long-term endeavor,” Facebook’s CTO Mike Schroepfer said in a post. “But if we can get them right we will be able to build systems that are smarter and more useful, enable developers to create immersive new experiences and make it possible to connect everyone in the world.”
Facebook’s Artificial Intelligence Research (FAIR) team plans to present a paper on its new system at the major AI conference NIPS next month. Schroepfer spoke about the project at the Dublin Web Summit on Tuesday.
Facebook also announced strides it has made in training machines to make inferences based on observations, a process known as unsupervised or predictive learning. FAIR said it has made a system that can “watch” a series of visual tests, such as a set of stacked blocks that may or may not fall, and predict the outcome. After a few months, according to the social network, the system can now predict the outcome 90% of the time, which is more accurate than most humans. The company is also working to teach systems how to plan. For example, Facebook said it created an AI bot that can play the board game Go as well as a skilled human player.
No comments:
Post a Comment