An AdaBoost Based - Deep Stochastic Configuration Network
Résumé
Deep stochastic configuration network (DSCN) is an incremental learning method for large-scale data analysis and processing, which has the advantages of lower human intervention, higher learning efficiency and stronger generalization ability. For improving the stability of DSCN, a deep stochastic configuration network based on AdaBoost is proposed, termed as AdaBoost-DSCN. In our proposed model, the AdaBoost learning approach is adopted, the weights of the base models are adjusted adaptively according to the training results, then the base models are combined to generate a stronger model, which is beneficial to reduce the influence of random parameters on network performance of DSCN. Experimental results on complex function approximation problems and large-scale regression datasets show that AdaBoost-DSCN has higher regression accuracy for large-scale data regression analysis compared with DSCN, AdaBoost-SCN, SCN.
Origine | Fichiers produits par l'(les) auteur(s) |
---|