The original article was written by Mike Schroepfer
Mike Schroepfer, the CTO of Facebook, reported in his blog the latest advancements of Facebook in the A.I. department. The Facebook Artificial Intelligence Research Team (FAIR) is currently working in the advancement of the A.I. learning patterns, specifically, the ability to train computers to identify objects, understand language, and plan.
Concerning to object detection, FAIR has found ways to significantly increase the efficiency of object detection in pictures. In a paper that will be published in December at the NIPS, FAIR presented the a system that supposedly diminishes the processing time of the images by 30% using 10 times less learning data than previous benchmarks with older methods.
In the natural visual understanding department, FAIR, earlier this year, showed the progress on their system Memory Network. The system Memory Network, previously, could only read and answer questions about a text, but now, it can also answer questions about a picture. In a video uploaded to the article, there’s an example of Memory Network in action. In the video, there’s a baby in the bathroom brushing his teeth. The user asked the system for the presence of the baby, the location of the baby, and what the baby was doing. Memory Network answered all the questions right.
FAIR has also been working in developing artificial predictive learning. The author defines predictive learning as “the ability to understand what will happen in the future by learning from observation.” To achieve this, FAIR has created a system that can predict if a bunch of piled boxes will fall or not. It is to be noted that the machine can only see the image and learn the patterns. After a few months of development, the system was reported to predict the outcome in 90% of the occasions, being much better than the average human.
If all the above doesn’t sound impressive enough, FAIR is also working in one of the most ambitious and difficult tasks of A.I., planning. To test some of their advancements in the planning area, FAIR created a machine that plays the game of GO. The machine learns the strategy on-the-fly, meaning that it doesn’t have a built in mechanism to play the game. After a couple of months of playing, the machine is on-par to a very good human player.
These advances are just the beginning of a very ambitious long-term plan of revolutionizing the technology industry by Facebook.