Fast sparse approximation of extreme learning machine

Xiaodong Li, Weijie Mao, Wei Jiang

Abstract: We introduce a fast sparse approximation schemes of extreme learning machine (ELM) named FSA-ELM of extreme learning machine (ELM). Our algorithms have two compelling features: low complexity and sparse solution. Experiments on benchmark data sets show that the proposed algorithm obtains sparse classifiers at a rather low complexity without sacrificing the generalization performance. As validated by the simulation results, FSA-ELM tends to have better scalability and achieves similar or much better generalization performance with much faster learning speed than the traditional ELM algorithm.