A couple developer friends and I formed a study group to go over all of the sessions in WWDC 2018. I reviewed all of the machine learning-focused sessions and gave an overview.
As most people in the group don’t have expertise in machine learning, I aimed at giving them a high-level view of Apple’s machine learning offerings on iOS, what is more likely to be relevant to ordinary iOS developers, and point to places where they might be interested to look into on their own.
Some of the take-aways I offered the group were:
Apple’s offerings expanded quite a bit to machine learning training this year, and if you want to do quick experimentations, it can sometimes make sense to train models on the Mac. However, for more serious applications, on-device inferencing using models produced from other existing workflows is probably more relevant.
There are some nice advantages to doing on-device inferencing. On iOS, there are some ML tasks that could leverage built-in frameworks and hardware, which is nice, although it would clearly be Apple-specific and not cross-platform.