Least Angle Regression Adaptive Incremental Broad Learning System
Fei Chu, Jiaming Su, Tao Liang, Junlong Chen, Xuesong Wang, Xiaoping Ma · 2020
Broad Learning System (BLS) is a latest proposed flat network, which has an effective and efficient learning capability. Due to excellent performance, BLS has been applied to many fields. However, there is little attention on the structure simplification of BLS network. In this paper, drawing lessons from application of Least Angle Regression (LARS) in feature selection, we propose Least Angle Regression Adaptive Incremental Broad Learning System (LARS-BLS). Begin with a network with complex structure, LARS-BLS firstly ranks all nodes, then, the nodes are added to the empty network in turn according to the contribution rate from high to low. Then, the adaptive coefficient is calculated according to the accuracy change rate of the model to adjust the number of nodes added to the model accordingly. When the modeling requirements are met, stop adding nodes. Finally, a sparse BLS model without redundant nodes is constructed. The proposed method is applied to a real-world application and some common data sets. The results show that our proposed algorithm is effective and efficient.