It may be that I've grossly misunderstood this article. I admit to scanning it because I'm quite busy today, but I'm intrigued with the idea of linking the effi

It may be that I've grossly misunderstood this article. I admit to scanning it because I'm quite busy today, but I'm intrigued with the idea of linking the efficiency of machine learning to the efficiency of human learning. It appears to me that the closest analog to the type of learning they are testing is spaced repetition. Interestingly every spaced repetition algorithm optimises at around 92% (mainly because every one is based on Super Memo's algorithm). I've long suspected that a lower value would be better -- my own spaced repetition algorithm uses 90%. Even if I've totally missed the point, I think it's an interesting idea to experiment with different values.