在线增量极限学习机及其性能研究

针对在线学习中极限学习机需要事先确定模型结构的问题，提出了兼顾数据增量和结构变化的在线极限学习机算法。算法于在线序列化极限学习机的基础上，通过误差变化判断是否新增节点，并利用分块矩阵的广义逆矩阵对新增节点后的模型进行更新，使模型保持较高正确率。通过在不同类型和大小的数据集上的实验表明，所提算法相较于经典极限学习机及其在线和增量学习版本都具有较好的分类和回归准确率，能够适应不同类型的数据分析任务。

Research on online incremental extreme learning machine and its performance

Aiming at the problem that Extreme Learning Machine (ELM) requires to determine the structure of the model in on-line learning, the paper proposed an on-line incremental ELM algorithm that supports both data increments and structural changes. On the basis of on-line sequential extreme ELM, the algorithm determines whether to add a new node or not by monitoring the changes of error. It uses generalized inverse of block matrix to help update the model after new node is included so that the model maintains high accuracy. Experiments on different datasets with diverse volumes show that, the proposed algorithm has better classification and regression accuracies compared with classic ELM and its on-line and incremental learning versions, and is suitable for various types of data analytical assignments.