Extreme learning machines with expectation kernels

作者：

Highlights：

• New insights to ELM from the perspective of kernel approximation us- ing random sampling techniques.

• Expectation kernel-based ELM (EKELM) outperforms ELM and KELM in terms of classification performance.

• Gaussian distribution to generate random weights, increasing the number of input weights will not cause overfitting to EKELM.

• Reducing the number of input weights by using proper kernel functions.

摘要

•New insights to ELM from the perspective of kernel approximation us- ing random sampling techniques.•Expectation kernel-based ELM (EKELM) outperforms ELM and KELM in terms of classification performance.•Gaussian distribution to generate random weights, increasing the number of input weights will not cause overfitting to EKELM.•Reducing the number of input weights by using proper kernel functions.

论文关键词：Bounded random feature mapping,Expectation kernel,Extreme learning machine,Kernel learning,Random sampling

论文评审过程：Received 13 September 2018, Revised 4 June 2019, Accepted 8 July 2019, Available online 23 July 2019, Version of Record 26 July 2019.

论文官网地址：https://doi.org/10.1016/j.patcog.2019.07.005