R. Kamimura and Y. Maruyama (Japan)
: mutual information maximization, competitive learning, cost, distinctive features, student survey
In this paper, we apply cost-sensitive information maxi mization to a student survey analysis. The cost-sensitive information maximization is developed to maximize infor mation while controlling a cost, that is, average distance be tween input patterns and connection weights. By introduc ing the cost in the framework of information maximization, we can flexibly control the feature detection performance of information maximization. We applied the method to a student survey analysis. Experimental results confirmed that information maximization and cost minimization can increase information to a sufficiently large level. We could see that cost minimization extracts the overall features of input patterns, while information maximization tries to de tect features distinctive of some input patterns.
Important Links:
Go Back
