It was a huge surprise for the technology industry, particularly the field of Artificial Intelligence as Apple announced that it will publish its first AI research paper, but what caught the industry off-guard is the fact that the Cupertino giant actually did.
It is not news to all how meticulous and confidential Apple is when it comes to its operations and creation process that is why, people were skeptical when the tech frontrunner announced its attempt to release its own publication.
However, the paper titled ‘Learning from Simulated and Unsupervised Images through Adversarial Training”, which was submitted for review in mid-November, was eventually published on December 22 through the Cornell University Library.
This is another breakthrough for the company who used to bar its employees from publishing researches related to artificial intelligence.
The research was authored by Ashish Shrivastava, a University of Maryland PhD holder in computer vision, together with a team of engineers that include Tomas Pfister, Oncel Tuzel, Wenda Wang, Russ Webb and Apple Director of Artificial Intelligence Research Josh Susskind, according to appleinsider.com Tuesday.
The paper focuses on techniques on how to train computer vision algorithms to distinguish objects by the use of computer generated, or synthetic, images.
However, although compared to training models solely based on real-world images, those utilizing synthetic data are often more competent because computer generated images are usually labeled, the gap between synthetic and real image distributions may be a factor in the performance of synthetic image learning.
To address this, Apple has proposed a Simulated plus Unsupervised (S+U) learning, where the task is learning a model to improve the realism of a simulator’s output using unlabeled real data while preserving the annotation information from the simulator.
The company has developed a method for S+U learning that uses an adversarial network similar to Generative Adversarial Networks (GANs), but with synthetic images as inputs instead of random vectors.
Although it does not necessarily foretell future consumer technology, one point of interest in this breakthrough is that Apple designated the application of its modified GAN in gaze and hands pose estimation evaluation. The company also added its interest in moving beyond static images to video input when it comes to S+U.
Interested in trading binary options. Don’t go anywhere else, try Options12.com.Options12.com brings your money back to you, the difference? Options12 prioritizes each client in maximizing his/her trading potential.
Financial trouble? Let Exo Capital Markets manage your funds accordingly. Exo Capital Markets strives to become the leading financial services firm by offering its clients with the most modern solutions in the industry.