Further experimental evidence against the utility of Occam's razor
Journal of Artificial Intelligence Research
Occam's razor just got sharper
IJCAI'07 Proceedings of the 20th international joint conference on Artifical intelligence
| Hi-index | 0.00 |
This paper examines a recent attempt to justify an inductive bias toward decision trees with few leaves. It is shown that this argument is invalid because it rests upon questionable assumptions, and can be used to deduce contradictory conclusions. Specifically, it can be used to prescribe any inductive bias. In general, it is shown that one cannot justify a preference for any inductive bias over another without making a priori assumptions about the distribution of target concepts. These results refute one common justification for Occam''s Razor, which recommends preferring simple hypotheses over complex ones when both are consistent with a set of observations.