Research of the Extreme Learning Machine as Incremental Learning

The external environment of the world is dynamically changing, which requires the ability of continuous learning and memorization from intelligent systems. Iterative learning methods neural networks have smooth convergence, so they do not allow achieving such an effect. An exception is the Extreme Learning Machine (ELM), which allows for incremental learning. In the ELM, the input weights are randomly generated. The weights between hidden and output layers are computed in one step, resulting in a faster learning rate compared to traditional SLFN learning algorithms. Incremental training allows the neural network to retain existing data while at the same time adapting to new data. This article presents the research and assessment of the ELM as incremental learning. In particular, research of the sample size effect for different numbers of neurons in the hidden layer on the ELM performance is presented.