Bidirectional Broad Learning System
PingQiang Huang, Bao Chen · 2020
As we all know, the learning efficiency and learning speed of traditional neural networks are far below requirements, which has become the main bottleneck of many applications. Recently, Chen et al. proposed a simple and effective learning method called Broad Learning System (BLS). This model experiment shows that this model can greatly reduce the training time compared with the traditional method. However, there are still many open questions in the field of BLS, such as whether the number of hidden layer nodes can be minimized without affecting the learning effect. This article proposes a new learning algorithm called the bidirectional learning system(B-BLS). In this algorithm, the feature layer nodes in the BLS are directly embedded by random projection, while the enhancement layer nodes are generated by random projection and the other half are obtained by reducing network residuals. Experiment results show the effectiveness of the method.