next up previous
Next: Learning by Taking Advice Up: Rote Learning Previous: Rote Learning

Store v Compute

Rote Learning must not decrease the efficiency of the system.

We be must able to decide whether it is worth storing the value in the first place.

Consider the case of multiplication -- it is quicker to recompute the product of two numbers rather than store a large multiplication table.

How can we decide?

Cost-benefit analysis
-- Decide when the information is first available whether it should be stored. An analysis could weigh up amount of storage required, cost of computation, likelihood of recall.
Selective forgetting
-- here we allow the information to be stored initially and decide later if we retain it. Clearly the frequency of reuse is a good measure. We could tag an object with its time of last use. If the cache memory is full and we wish to add a new item we remove the least recently used object. Variations could include some form of cost-benefit analysis to decide if the object should be removed.



dave@cs.cf.ac.uk