Quote:
Originally Posted by zhou_jinyuan
I have confusion too. As I understand hypothesis set is associated with a learning algorithm. does g in choice from a to d come from same learning algorithm or description represents different algorithm? Since we have 256 possible hypothesis, I can conceptually call my learning algorithm "try all" which have all 256 possible functions as its hypothesis. Does this exercise assume we are working with "try all" algorithm?

A hypothesis set is just that; a set of hypotheses. The algorithm is a separate entity that chooses the final hypothesis from this set. It can in principle make that choice any way it wants (some algorithms may be better than others for the same hypothesis set).
To answer your question, the algorithm can try all hypotheses (in the hypothesis set), but it will have to choose one and only one as the final hypothesis that it reports. When we grade the algorithm, what matters is the performance of the final hypothesis it arrived at, regardless of how it arrived at it.