Apple recently launched a publishing site for its machine learning research to bring its findings on machine learning to the masses. As of now, the Apple Machine Learning Journal has only one paper published, but more posts are expected in the coming weeks. The move by the electronic giant is considered as a new step, as the firm was tight-lipped about its projects and researches until the date.
The blog and its initial post reveal some interesting facts. The reports confirm that the initial research paper was already published on arXiv, an online preprinting repository for publishing papers. It is also concluded that Apple is trying to attract the research talents through the blog as the initial post also contained a link to Apple jobs page at the bottom. Industry experts say that Apple is trying to grab the attention of machine learning through the initiative as most people think Google and Amazon are doing a better job in machine learning compared to the smartphone major.
The research paper published on the platform discusses the possibilities of improving the realism of various synthetic images. The blog discusses different methods applied to improve the accuracy on various machine learning tasks using the refined images. For illustrating the realism of images better, the blog added images, GIFs, and more. It also discusses a unique and inexpensive method developed by Apple to improve the realism of artificial images.
The paper says that the firm had to train its neural networks to identify the faces and other objects in various photos. The company made a collection of synthetic images to train the neural networks instead of going for any image data base of real images. Interestingly, the tech giant applied its unique method to make them more real and claims that the technique offers faster and cheaper option to train the neural network.