This interesting new paper tries to improve on model-based SRS scheduling algorithms by adding *planning*—that is, given a student's time constraints, find an optimal sequence. Surprisingly, doesn't do significantly better than typical "greedy" approaches. arxiv.org/pdf/2102.04174
Conversation
I guess I find that encouraging! A simple "schedule everything below a fixed predicted-recall threshold" model does as well as holistic planning optimization, (assuming you're already using a sophisticated model for predicting recall rates). Much more flexible this way.
Replying to
Errant (unrelated) thought: recall for prompts in the mnemonic medium is so high because there's a surrounding text, forming a natural elaborative encoding.
1
Yes, I think something like this is likely.