dewcon is exactly right.
My opinion is that the learn function -- unless you are hunting for shallow coins and don't mind missing a bunch of them that don't ID correctly -- is a waste of time.
There are umpteen different situations, ground conditions, target locations and orientations in the ground, proximity to other targets, etc. etc. etc. that will drastically alter the way a coin IDs. In general, the trashier the ground, the worse the IDs, and the deeper the target, the worse the IDs. The idea of the "learn" function is simply that it takes a small block of target IDs and "accepts" them. But, if for whatever reason the coin you are searching for falls outside of that small block of target IDs, then you will NOT detect that coin.
If you are simply wanting to fly through a park and cherry pick, say, shallow dimes and quarters, then using the "learn" function to ignore all other targets will allow you to very quickly dig some shallow dimes and quarters. But, in any other scenario, I think the "learn" function would be more of a hindrance than a help. In my limited experience, I believe that to begin making more and better finds with the Explorer, you must accept smoe deviation away from "proper" ID numbers or specific cursor positions, understanding that there are countless reasons why a given target will not ID "properly."
Sorry for the long-winded answer -- but my best advice for anyone is to learn to deviate away from "perfect" ID if you want to dig deeper and better finds. That's kind of contrary to the idea of the "learn" function...
Steve