我的新闻

Broad Learning System (BLS) is a recently proposed single-layer feedforward network (SLFN) with strong generalization ability in different industrial applications. However, the classical BLS assumes that the training and testing data are drawn from the same distribution, which can be often violated in the real world. This paper proposes a novel dynamic domain adaptation (DA) framework based on BLS (DDA-BLS). Compared to most existing DA methods, which follow a static feature learning protocol, the proposed DDA-BLS applies a data-dependent dynamic feature learning procedure for different target inputs. Although such a dynamic feature learning procedure seems to be a more intelligent DA strategy and improves DA performance, it has rarely been explored in the DA fields. Comprehensive experiments on several DA tasks, including image classification and fault diagnosis, demonstrate the effectiveness and efficiency of the proposed DDA-BLS in DA tasks, further indicating the superiority of the dynamic DA strategy.