The learning model is a set of hypothesis {g_a}, one of which is, hopefully, something like the target function.