You can remember past observations as a shortest program that matches them, potentially compressing them fairly dramatically. If it still doesn’t fit, one could probably also perform some sort of search for the optimal lossy encoding scheme as well. This would break your assumptions about length of remembered s being independent from the hypotheses.
You can remember past observations as a shortest program that matches them, potentially compressing them fairly dramatically. If it still doesn’t fit, one could probably also perform some sort of search for the optimal lossy encoding scheme as well. This would break your assumptions about length of remembered s being independent from the hypotheses.