A team of Apple employees has published a research paper on artificial intelligence. While the topic is, in and of itself, interesting and a hot one right now, this paper is particularly notable because it’s the first one published by an Apple team since the company started allowing its employees to publish research papers.
Apple studies artificial intelligence
The paper is entitled “Learning from Simulated and Unsupervised Images through Adversarial Training, and it was submitted for publication on Nov. 15. It was released on the arXiv server last week. Ashish Shrivastava and a time wrote the paper and submitted it within weeks of Apple’s announcement that it would begin allowing its employees to publish research papers.
Baupost's investment process involves "never-ending" gleaning of facts to help support investment ideas Seth Klarman writes in his end-of-year letter to investors. In the letter, a copy of which ValueWalk has been able to review, the value investor describes the Baupost Group's process to identify ideas and answer the most critical questions about its potential Read More
The Apple team focused on technology dealing with intelligent image recognition, an area that it’s no surprise that the company is studying. The researchers describe a way to make a computer program able to recognize images that were generated by a computer and also decipher them. Thus far, the main issue has been that images that are generated by computers are usually not realistic enough for a program to be able to recognize them.
Some progress made
Another big problem, according to the Apple researchers, is that it is “computationally expensive” to make those computer-generated images realistic enough for a computer program to be able to decipher them. Their technique appears to offer some improvement, as recent progress in the area of graphics has made it possible for them to train some models to recognize images generated by computers.
However, the researchers note that there is still something to be desired in terms of “learning from synthetic images” because it “may not achieve the desired performance due to a gap between synthetic and real image distributions. In order to move another step toward image recognition, they suggest their technique called “Simulated+Unsupervised (S+U) learning.” The aim of their technique is to make computer-generated images more realistic with the use of “unlabeled real data.”
To do this, they used what’s called “adversarial learning,” in which a competition between neural networks is set up.
Apple’s actually sharing
CNET‘s Jessica Dolcourt feels that the technique described in the paper isn’t necessarily the big takeaway here. She notes that this paper represents the opening of Apple’s research vault. The company is widely known for its strict secretiveness, as it has long been unwilling to share anything it has learned with anyone else. However, other tech firms have begun to share their technologies with each other, like Facebook’s open source VR camera. Tesla also opened up its patents for others to use in hopes of accelerating adoption of electric vehicles.
Artificial intelligence has been an area of interest at Apple for some time, as it is the basis of Siri. However, some analysts said earlier this month that it will be Microsoft, Amazon and IBM that will lead the way in artificial intelligence in 2017—not Apple.