Incremental Shared Subspace Learning for Multi-label Classification

Abstract

Multi-label classification plays an increasingly significant role in most applications, such as semantic scene classification. In order to exploit the related information hidden in different labels which is crucial for lots of applications, it is essential to extract a latent structure shared among different labels. This paper presents an incremental approach for extracting a shared subspace on dynamic dataset. With the incremental lossless matrix factorization, the proposed algorithm can be incrementally performed without using original existing input data so that to avoid high computational complexity and decreasing the predictive performance. Experimental results demonstrate that the proposed approach is much more efficient than the non-incremental methods.