Modeling Active, Hypothesis-Driven Learning From Worked-Out Examples