Apple’s machine learning advances
You’ve probably heard about machine learning – but what is it, and how is it changing your Apple kit?
The phrase ‘machine learning’ conjures up images of futuristic tech that wouldn’t look out of place next to Sarah Connor. But despite its recent emergence at Apple events and in the high-tech press, it’s actually been around for a long time. The way an email app’s spam-detection function improves over time is one example. A computer game learning the most effective way to get a high score is another.
So what is machine learning, and how is it different from similar terms, such as artificial intelligence? Put simply, machine learning is where a computer learns how to solve a problem using copious amounts of data. The computer learns this not by relying on code created by a human for that purpose, but by learning from previous examples.
There are certain crossovers with data mining, but they are different. Both utilise huge amounts of information, but data mining involves scouring that data for certain matches (creating a list of Apple Stores in the UK, for example), while machine learning involves teaching a machine something new based on that information. Artificial intelligence, on the other hand, is an umbrella term for machines being able to carry out tasks that would be considered intelligent were they done by a human.
Apple has made much of its products’ machine learning capabilities in recent years. You may have heard it boast of the iPhone’s Siri Shortcuts or augmented reality talents, or how the QuickType keyboard utilises this tech to give you better suggestions as you type. All of these features rely on machine learning to not only do what they do, but to improve over time as well.
This all sounds very in-depth and potentially confusing, so how will it actually benefit you? Apple wants it all to appear very much under the hood – if you didn’t pay attention to Apple’s machine learning advancements at its recent
events, you may not even know about them. All the complicated computational learning is going on behind the scenes, creating systems and processes that appear simple on the surface.
Playing phone tag
Take the tagging capability of the Photos app, for example. In the past you’d have to remember to tag all your photos appropriately if you wanted to find them in search results. If you forgot, you’d have a choice between tagging potentially thousands of pictures or simply ignoring the tagging feature altogether. Now, all of that is done by the iPhone itself, with no input required from you at all.
Your device learns what is in each photo – and gets progressively better at doing so – with the end result being that your photos are essentially auto-tagged by your iPhone, so you can search for ‘beach’ or ‘dog’, say. Here, complex behind-thescenes tech simplifies the process for you.
Here’s another example. On 30 July 2014, Apple switched Siri’s voice recognition to a machine learning-based system in an attempt to improve its accuracy. “The error rate has been cut by a factor of two in all the languages, more than a factor of two in many cases,” said Alex Acero, Siri senior director at Apple. And it’s not only error reduction in Siri that has benefitted from machine learning. Before you invoke the digital assistant, it’s listening out for the “Hey Siri” trigger phrase. A deep neural network converts the audio signals from your voice into a probability score to decide whether you did in fact utter the trigger phrase. If it believes that you did, Siri starts up.
The Apple Pencil also relies on machine learning in its ‘palm rejection’ technology,