基于截断最大相关熵准则损失函数的鲁棒极限学习机

Robust Extreme Learning Machine Based on Truncated Maximum Correlation Entropy Criterion Loss Function

摘要: 极限学习机(Extreme Learning Machine, ELM)在前馈神经网络中具有高效快速的优势。然而ELM在面对异常值时较敏感，鲁棒性较差。本文对最大相关熵损失函数进行了截断，将最大相关熵损失函数进一步改进为能够限定最大损失为常数的截断最大相关熵损失函数，构建基于截断最大相关熵准则损失函数的鲁棒ELM模型，以此来抑制噪声和异常值对模型的影响。采用迭代重加权算法对模型进行求解。最后在数据集上验证所提出的模型的有效性。实验结果表明，该模型在噪声条件下提高了对噪声和离群值的鲁棒性。

Abstract: Extreme Learning Machine (ELM) has the advantage of being efficient and fast in feed forward neu-ral networks. However, ELM is sensitive to outliers and has poor robustness. In this paper, the maximum correlation entropy loss function is truncated, and the maximum correlation entropy loss function is further improved to a truncated maximum correlation entropy loss function that can limit the maximum loss as a constant. A robust ELM model based on the truncated maximum cor-relation entropy criterion loss function is constructed to suppress the influence of noise and outliers on the model. The iterative reweighting algorithm was used to solve the model. Finally, the validity of the proposed model is verified on the data set. Experimental results show that the robustness of the model to noise and outliers is improved under noisy conditions.