CAJ | 학술논문

基于传统的神经网络理论，并结合原始的ELM 算法，提出一种新的确定网络结构的方法，即二分法的极端学习机。该算法无需人为调整任何参数，整个训练过程都是全自动的，这为今后寻求更好地确定网络结构的方法提供了一个好的思路。数值实验结果表明，该方法不仅收敛速度快且能获得较紧凑的网络结构，同时还具有较好的泛化能力。
기우전통적신경망락이론，병결합원시적ELM 산법，제출일충신적학정망락결구적방법，즉이분법적겁단학습궤。해산법무수인위조정임하삼수，정개훈련과정도시전자동적，저위금후심구경호지학정망락결구적방법제공료일개호적사로。수치실험결과표명，해방법불부수렴속도쾌차능획득교긴주적망락결구，동시환구유교호적범화능력。
A new method to determine the networks is proposed .Based on the traditional neural network theory ,combining the previous ELM algorithm ,a new method to determine the network structure which called the bisection ELM is put forward .This new algorithm is full-automatically to determine the net-work architecture which needs not to adjust any parameters ,and it gives a good idea to find better meth-od in the future .Simulation results demonstrate and verify this new method not only achieve extremely fast convergence rate and obtain a more compact network structure ,but also have good generalization performance .