Distributed Broad Learning System
Yifan Zhai, Ying Liu · 2020
Recently, broad learning system (BLS) has been proposed and widely applied to the fields of machine learning and time series analysis. Compared with the well-known deep learning models, BLS does not need the deep architecture, and the learning process is time efficient. However, the existing BLS belongs to centralized processing, which is not applicable to the cases when data are distributed over multiple nodes. To solve this problem, in this paper, we propose a consensus-based distributed implementation of BLS (dBLS), in which each node cooperates with its one-hop neighbors to train the weights of the dBLS. Besides, a distributed extreme learning machine auto-encoder (dELM-AE) is also developed to refine the features extracted from the input data. Some simulations are performed and results show that the proposed dBLS is effective in solving distributed classification problems.