Apple Launches Machine Learning Journal ⇥ macstories.net

Ryan Christoffel, MacStories:

Apple has historically been secretive when it comes to its machine learning research, a stance which many speculated had put the company at a disadvantage in the area of recruiting talent. It’s understandably hard to build a noteworthy reputation as a machine learning researcher if you’re unable to talk about any of your work. But near the end of last year Apple’s director of AI research, Russ Salakhutdinov, signaled that change would be coming to the company’s policies surrounding secrecy. The launch of a public journal featured on Apple’s website is very clear evidence of that change arriving.

Apple’s journal probably shouldn’t necessarily be considered a “journal” in the academic sense — there’s no authorship information, and it isn’t peer-reviewed. It’s more like a journal in a synonym-for-a-diary sense, and there’s nothing wrong with that. If it’s updated regularly, it will provide a glimpse into the ways Apple is developing machine learning technologies in a way that should be beneficial to researchers within the company and those who are working on it in other capacities.

For some reason, the virtually all-text site includes on every page a 1 MB JavaScript file. The first published article is a 4 MB page which, while not an egregious file size generally, is pretty ridiculous for something that’s largely text. A great deal of the additional bulk comes in the form of images that could easily be compressed: the lead image is a 110 KB PNG file, which I losslessly compressed to 30 KB by running it through a basic image optimization tool. It’s a little thing, but the little things stack up to a text-based article that wastes multiple megabytes of bandwidth.