Apple’s First Published AI Research Paper Is Focused on Improved Computer Vision

Apple’s First Published AI Research Paper Is Focused on Improved Computer Vision
Text Size
- +

Toggle Dark Mode

Apple’s first AI research paper — published on Dec. 22 — describes a technique to improve the recognition abilities of computer vision systems.

The paper, “Learning from Simulated and Unsupervised Images through Adversarial Training,” was submitted for review in mid-November, and was published by the Cornell University Library. Its lead author is Ashish Shrivastava, with co-authors Tomas Pfister, Oncel Tuzel, Wenda Wang, Russ Webb, and Josh Susskind, and was first spotted by Forbes on Monday.

Cupertino’s first published research paper goes into detail about machine learning computer vision. For example, it’s typically more efficient to use synthetic image data because it comes labeled and annotated — unlike real-world image data, which requires human operators to painstakingly label everything within the image. Unfortunately, using synthetic image data has its own set of problems, as it’s often “not realistic enough, leading the network to learning details only present in synthetic images and fail to generalize well on real images,” according to the paper.

To solve this, the research paper describes a method, called Simulated+Unsupervised learning, to boost the realism of synthetic images. To do so, Apple’s researchers said that they are using a version of a machine learning technique called Generative Adversarial Networks, which takes two neural networks and pits them against each other. The result is more photorealistic images, CIOL reported. The end result of the research could be machine-learning algorithms that are better at recognizing objects and traits within images. In the future, the team hopes to apply the method to videos.

The paper’s publication comes less than a month after Apple announced that it would allow its artificial intelligence employees to publish their research and confer with other researchers. Previously, Cupertino barred its researchers from publishing due to the possibility of losing commercial intellectual property, according to AppleInsider.

Additionally, the paper could be an important step for Apple’s future research. Cupertino’s previous research secrecy has historically hurt the company’s recruiting efforts, as publication and presence are key factors to many promising graduates, Bloomberg reported.

Social Sharing