基于l_(0)范数的鲁棒极限学习机的稀疏算法研究

Sparse algorithm of robust extreme learning machine with l_(0)-norm

为了进一步提高极限学习机(extreme learning machine,ELM)的稳定性和稀疏性,在鲁棒ELM的基础上,引入l_(0)范数作为模型的正则项来提高稀疏性,建立了基于l_(0)范数正则项的稀疏鲁棒ELM。首先,通过一个凸差(difference of convex,DC)函数逼近l_(0)范数,得到一个DC规划的优化问题;然后,采用DC算法进行求解;最后,在人工数据集和基准数据集上进行实验。实验结果表明:基于l_(0)范数的鲁棒ELM能够同时实现稀疏性和鲁棒性的提升,尤其在稀疏性上表现出较大的优势。

Extreme learning machine(ELM)has shown great potential in machine learning because of its high learning rate and strong generalization performance.In order to improve the robustness and sparsity of ELM,on the basis of robust ELM,l_(0)-norm is introduced as the regularization of the model to improve the sparsity,and a sparse and robust ELM based on l_(0)-norm regularization is established.A difference of convex(DC)function is used to approximate the l_(0)-norm.The optimization result is expressed as a DC programming,which is then solved by DC algorithm.Experiments on artificial and benchmark data sets show that the robust ELM based on l_(0)-norm can be improved in both sparsity and robustness,especially in sparsity.