Distributed sparse learning for stochastic configuration networks via alternating direction method of multipliers

As a class of randomized learning algorithms, stochastic configuration networks (SCNs) have demonstrated excellent capabilities in various intelligent scenarios. Regarding these algorithms, a growing interest in recent years has been shown in sparse representation, which is a powerful tool in many large-scale applications with high-dimensional data. However, few methods currently exist to fit sparse models in distributed environments. In this paper, we focus on devising two fully distributed learning algorithms for SCNs with sparsity regularization. The first algorithm is based on sample-wise division which is appropriate for the problem of large samples. The second algorithm is fit for high-dimensional data, which is based on feature-wise division. The sparse solutions for the decentralized optimization problems are obtained by introducing the alternating direction method of multipliers (ADMM) and the distributed average consensus (DAC) algorithm. Experimental results are provided to demonstrate the effectiveness of the proposed algorithms.