All the technology giants like Facebook, Google and Microsoft etc. have done hundreds of publications but Apple never published a paper keeping its magical codes etc. to itself. Just like Android is open to all devices but operating system of Apple is still unknown to be copied. Now, Apple Jumps into AI Research with Improved Stimulated and Unsupervised Image Through Adversarial Training.
But now the things have started changing as Apple director of AI Research, Russ Salakhutdinov has announced that they will be soon publishing their research papers. It seems Apple is dealing with things sensibly and timely now.
Apple Jumps into AI Research with Improved Stimulated and Unsupervised Image Through Adversarial Training
Six researches from Apple have recently published a paper that describes the methods for simulated + unsupervised learning. The basic reason is behind this publication is to improve the quality of synthetic images. This new step shows the company’s interest to become more visible leader in growing field of AI market.
Now a days machine learning models are trained by using synthetic images and videos of very high frequencies. Now days the images generated are less costly and more effective. This synthetic training is easy but risky as small error would lead to worse negative effects on final result. In other words it would be difficult to ensure that the generated image is of same quality to that of original image.
Now to improve the quality of these synthesized images apple has decided to use GANs (Generative Adversarial Networks). GANs is an old technology but Apple has revealed to use it with certain modifications.
The researchers working on this publication also did some modification. It includes the models to use full history of redefined images. The main reason behind this is to make network recognize that the image generated is real or fake. To read more about the work click here.