Apple may be keeping mum about its latest research and innovations in consumer technology, but the company is definitely taking a different approach as they venture into artificial intelligence (AI).
Earlier this month, Apple hinted at the possibility of the company openly publishing its AI research via a tweet posted by AI director, Russ Salkhutdinov:
— hardmaru (@hardmaru) December 6, 2016
Apple researchers have now released their first AI paper focusing on computer vision technology—specifically, teaching AI to recognize objects using simulated images.
The paper, titled “Learning from Simulated and Unsupervised Images through Adversarial Training,” describes how their program can identify and understand digital images by training computer vision algorithms to recognize synthetic 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. We show that this enables generation of highly realistic images, which we demonstrate both qualitatively and with a user study.
Basically, training models that use computer generated images are easy to identify because they are tagged; as opposed to real-world images which aren’t. A synthetic image of an eye could be easily identified by the algorithm, but a similar image taken from the real-world would be unknown to the algorithm because it isn’t labelled.
The paper suggests pitting neural networks against each other to bridge this gap—training the program so that it can better discern synthetic data from real data.
Perhaps even more relevant than the the technique described in the paper is the fact that it was even published at all. Apple is notoriously secretive about its research, and AI is the heart of some of its biggest innovations, like Siri.
The company’s reluctance however, may have proven to be counterintuitive to their progress in the field. Their intent to protect proprietary research and reluctance to share knowledge in AI may have turned off prospective talents who wanted their achievements recognized.
Nevertheless, their willingness to share information and research is a welcome one for a very competitive industry that is rapidly advancing. And a more democratized approach to the study could pave the way for more talent and brains working towards improving the technology.