Ensemble of Online Extreme Learning Machine with Progressive Amnesia

摘要

In recent years, Extreme Learning Machine attracts much attention because of its quickness and simplicity. However, the sample data is episodic and consistent so that they can't be trained at the same time, and be ballooning numbers. On the other hand, the older data also become due sometimes. These reasons cause to the training proceeds becoming more complex. To solve the problem, this paper introduces a progressive amnesia mechanism which helps strengthen the recent data but weaken the older data. The method is to construct a failure function for segmenting data. Based on this mechanism, a improving Extreme Learning Machine, which is named as Ensemble of Online Extreme Learning Machine with Progressive Amnesia, is shown. The experimental results demonstrate that the paper's method can effectively control the expansion of data and improve the forecast accuracy.