Apple’s latest machine learning breakthrough could help train better AI


Apple's secret AI sauce gets a new ingredient
Apple is working hard to improve its AI technology.
Photo: Charlie Sorrel/Cult of Mac

Having lagged behind in artificial intelligence for several years, Apple is eager to prove that it’s catching up.

In a new paper, published in Apple’s own Machine Learning Journal, it describes an approach to “unsupervised domain adaptation.” It claims that this can help improve the performance of “deep learning” models in certain scenarios.

In the paper’s abstract, Apple notes how:

“Deep neural networks are a milestone technique in the advancement of modern machine perception systems. However, in spite of the exceptional learning capacity and improved generalizability, these neural models still suffer from poor transferability. This is the challenge of domain shift — a shift in the relationship between data collected across different domains (e.g., computer generated vs. captured by real cameras). Models trained on data collected in one domain generally have poor accuracy on other domains.

In this article, we discuss a new domain adaptation process that takes advantage of task-specific decision boundaries and the Wasserstein metric to bridge the domain gap, allowing the effective transfer of knowledge from one domain to another. As an additional advantage, this process is completely unsupervised, i.e., there is no need for new domain data to have labels or annotations.”

You can read the paper here. It’s not clear exactly how this research will be used by Apple. However, it could be used for Apple’s Street View-style mapping initiatives or self-driving car project. As Apple notes, “Our method produces cleaner predictions and less confusion between challenging classes, such as road, car, sidewalk, and vegetation.”

Upping its AI game

For a long time, Apple separated itself from the rest of the A.I. community. It refused to attend conferences, and didn’t let its researchers publish their work in academic journals. Launching its quasi-academic blog, the Apple Machine Learning Journal, was something of a compromise.

When it launched back in 2017, Apple described it as a way to talk about how, “using machine learning technologies [can] help build innovative products for millions of people around the world.”

Apple’s AI efforts ramped up earlier this year when it hired Ian Goodfellow as its new director of machine learning. The former Google AI created general adversarial networks (GANs), an important development in modern AI.