Apple said at the beginning of the month that it will publish their AI research results publicly. The first paper was also unveiled a few days ago, the theme is the "visual identification" of computers.
The paper, entitled "Learning from Simulated and Unsupervised Images through Adversarial Training," explains Apple's view of image recognition. Traditionally, letting a computer "learn" to identify an image, usually using a composite image such as a video game, rather than a real photo, because the composite image is pre-populated with a label, such as "dog." In contrast, looking directly at the real image, the computer needs to manually re-label the objects in the photo, and then judge the "dog" from a background. This will result in a much lower efficiency of identification.
However, there is also a disadvantage in this way. Since the pictures that the computer usually "learns" are synthetic images, but the real photos are taken after the identification, the effect of this type of image recognition algorithm is not satisfactory. Therefore, the approach that Apple researchers have come up with in this paper is to make the composite image closer to reality.
This method first establishes two sets of AI neural networks, one is responsible for data input (generator), and the other is a discriminator, which is responsible for identifying the data produced by the generator, as well as the general "learning" image, and finally Let the two groups of AI fight each other, forcing closer to the real image. In fact, such data processing methods are the basic methods of GeneraTIve Adversarial Networks (GANs).
Interestingly, some of the researchers are members of startups acquired by Apple, and some are from universities. For example, the co-author of the paper, Joss Susskind, is the founder of EmoTIent. Their technique is to recognize people's emotions through expressions through AI, but was acquired by Apple in January this year. Apple's researchers also said they hope to move from the picture to the film identification.
This paper was published by Cornell University. In the announcement of Apple's recent announcement, Apple also opened its researcher or scholars, and was free to submit their own AI research results, contrary to the tradition of secrecy.
Note: Due to the public release, there should be no idea of ​​introducing commercial products in the short term.Piezoelectric Motor,Piezoelectric Vibrating Motor, Piezo Motor
NINGBO SANCO ELECTRONICS CO., LTD. , https://www.sancobuzzer.com