| Summary: | 碩士 === 國立中興大學 === 資訊科學研究所 === 82 ===
Machine learning is one of the most important subjects in Artificial Intelligence since learning is the core of wisdom. Learning from examples are widely studied and are applied in various areas. Most of the known learning algorithms classify concepts according to the attributes. This approach has some disadvantages, especially in situations of uncertainty and noise. They will make the these algo thms unable to work or inaccurate.
In this thesis, we propose another approach: learning concepts via learning the relative importance of attributes. Our method is to give higher weights to more important attributes. Because it is very hard to decide the important attributes among several concepts, the approach we take is to compare each pair of concepts, and use the vote heuristic to decide the concept. This model provides a solution for dealing with data uncertainty and noise. Since the absent attributes are always unimportant, and the ratio of noise data must be very low, they will not have much effect.
|
|---|
