Block-sparsity recovery via recurrent neural network

Several researches on block-sparsity recovery have been recently carried. Block-sparse signals with nonzero elements occurring in clusters arise naturally in many practical applications. However, the priori knowledge of block partitions is usually unavailable in reality, which increases the difficulty of recovery greatly. At the meantime, deep learning methods have been developed rapidly in last few years as a kind of data-driven methods. In this paper, we propose a novel approach based on recurrent neural networks for recovery of block-sparse signals with unknown cluster patterns. In our work, the recurrent neural network containing the long short-term memory is introduced to acquire the spatial correlations between nonzero elements of block-sparse signals. Extensive experiments both on synthetic data and real-world data show that the proposed method outperforms the state-of-the-art algorithms.