Earlier this month news broke stating that Apple will allow its researchers to publish AI papers. Despite Apple being a secretive company, Apple has released its first AI paper which focuses on Apple’s progress in the intelligent image recognition field.
The paper titled “Learning from Simulated and Unsupervised Images through Adversarial Training,” describes a program to improve the training of an algorithm’s ability to recognize objects using computer generated images or “synthetic” images, this is a more advance technology than the facial recognition features used in iOS 10 Photos app.
While real images require annotations to be added to them which is considered an “expensive and time-consuming task” since someone has to label each objects in a picture, the synthetic images make this processes easier “because the annotations are automatically available.” However, synthetic images could lead to a problem in the quality of the program. The synthetic images won’t be able to produce real-world images since it “is often not realistic enough, leading the network to learnt details only present in synthetic images and fail to generalize well on real images,”.
The paper suggests to improve synthetic image data, they will propose Simulated+Unsupervised (S+U) learning, this will allow a stimulated image to be boosted. The researchers also use a technology called Generative Adversarial Networks which puts two neural networks against each other and has been used to generate realistic images.
“In this paper, we propose Simulated+Unsupervised (S+U) learning, where the goal is to improve the realism of synthetic images from a simulator using unlabeled real data. The improved realism enables the training of better machine learning models on large datasets without any data collection or human annotation effort.”
The lead author of the paper is Apple’s researcher Ashish Shrivastava, the other Apple employees who co-authored on the paper include Tomas Pfisher, Oncel Truzel, Josh Susskind, Wenda Wang, and Russ Webb. The team’s first research paper was submitted on November 15, but wasn’t published until December 22.
This is a huge step for Apple, by publishing its research paper helps push for more advance AI software in the Tech industry. As time progresses, we should see a variety of researcher papers from Apple.
Source info: Forbes