Multi-label extreme learning machine based on label matrix factorization
Sihao Li, Chen Fucai, Ruiyang Huang, Xie Yixi · 2017
Multi-label learning aims to predict the label sets an instance belongs to. Extreme Learning Machine (ELM), as a single-hidden layer feedforward neural network algorithm, has been extended to multi-label scenarios because of faster learning speed and less human intervention. Aiming at dealing with the problem of ignoring the inter-label dependencies, the proposed method, ELM-LMF, decomposes the label matrix into latent label matrix mapping the label space to the latent space and the k-label dependency matrix preserving the inter-label relationships, classifies in latent space with ELM, and maps the predicted labels back to the original space. Experiments on 3 benchmark multi-label data sets with 3 other state-of-art methods prove the feasibility of ELM-LMF.