Abstract:A major assumption in many machine learning algorithms is that the training data and testing data have the same distribution. However, in many real-world applications, this assumption may not hold. Transfer learning addresses this problem and utilizes plenty of labeded data in a source domain to solve related but different problems in a target domain. This paper proposed a parameter- transfer learning method based on KMM (Kernel Mean Matching) algorithm. First, we weighed each source instance using KMM and then applied the reweighted instances to the learning method based on parameters. We applied this method to the localization of wireless network. Experiment results have demonstrated that the proposed method outperforms the methods based on instances or parameters, especially when the target training data are relatively few.

Abstract:Conventional method of error compensation for truck scales is fussy and the weighing results are not precise, so an error compensation method based on multiple radial basis function neural networks (RBFNN) was proposed. The sub-RBFNNs were confirmed according to the maximum permissible error of scales different verified weighing capacities. The compensation models were established and then the algorithm of training RBFNN was presented. Sub-RBFNNs were combined and one of them was chosen to fully compensate weighing error of every capacity by adaptive network, and then optimum compensation of total weighing range was obtained. Simulation experiments show that the sub-RBFNNs size of this proposed method is smaller and the errors are less than those of the method that compensates the error of total weighing range with single RBFNN.