Idle thought: could I update the spaced repetition algorithm that my flashcard app (Fresh Cards) uses to make use of CoreML for machine learning? Fresh Cards records all your card review evaluations, so it should be possible to process all the data about how well you recall all cards over time to adjust the repetition intervals. The goal would be to use CoreML to do a sort of curve-fit of your recall data. The curve would be the Ebbinghaus forgetting curve and the data points would be your review timestamps and review scores.
Some interesting ideas here… More to come.