Broad learning system based on Elastic Net feature sparsity and dense
Haofeng Ding, Shiwen Xie, Yongfang Xie, Jie Wang · 2021 China Automation Congress (CAC) · 2021
Broad learning has attracted wide attention since it was proposed, because of its characteristics of short training time and online updating. However, there are many problems in the original broad learning, such as low prediction accuracy, large computational consumption of resources, and many redundant data in the calculation process. Therefore, this paper proposes a broad improvement method based on dense connection of feature layers and using Elastic Net to sparse feature layers. The different Windows of the feature layer of the original broad learning system were put into the form of dense connection, and then Elastic Net was introduced to sparse the feature nodes, and then the sparse feature nodes were combined with the input data as the input of the enhancement layer nodes, making the model more compact. The proposed method was applied to a public dataset to verily the results. The experimental results show that the proposed method has short training time and higher accuracy than other methods.