Networking Of Generalized Approximate Message Passing Algorithm Based On Deep Learning

Generalized Approximation Message Passing algorithm,Bilinear Generalized Approximation Message Passing algorithm and Parametric Bilinear Generalized Approximation Message Passing algorithm have a wide range of applications on Conpressed Senssing.Based on Deep Learning techniques,this paper focuses on networking these three popular algorithms by Tensorflow.The data set is used to update parameters in training mode,reducing the prior knowledge required by the algorithm.Make the parameters more suitable for the solution of the target problem,and show better performance in specific applications.The specific content of the paper is as follows:Firstly,introduced the related knowledge of deep network in Deep Learning,and networked the Generalized Approximate Message Passing algorithm.Set the learnable parameters according to the inter-layer variables of the Generalized Approximate Message Passing algorithm network.In the learning training,the Adam optimizer is used to continuously optimize and update the network parameters through a large amount of sample data,so that the network performance is continuously improved.The experiment proves that the corresponding algorithm network is constructed for the signal reconstruction problem,so that the network can train the update parameters according to the sample data,reduce the dependence on the prior knowledge,and improve the signal reconstruction performance.Then,networked the Bilinear Generalized Approximate Message Passing algorithm,and the neural network structure of Bilinear Generalized Approximation Message Passing algorithm is constructed.The sampled data and optimizer optimize the learnable parameters to reduce the matrix and dependence of prior knowledge of the original signal.Experiments show that after the introduction of deep learning,the networked Bilinear Generalized Approximation Message Passing algorithm is greatly improved on the Dictionary Learning problem compared with the original algorithm.Finally,networked the Parametric Bilinear Generalized Approximation Message Passing algorithm,the algorithm structure is analyzed,and the deep network corresponding to the algorithm is designed to reduce the number of parameters that can be learned in the Bilinear Generalized Approximation Message Passing algorithm.And use the calculation method of Kronecker product to participate in network construction,reduce the computational complexity,optimize the network according to data and optimizer,reduce the dependence of the algorithm on prior knowledge,and Parametric Bilinear Generalized Approximation Message Passing algorithm networking effectively improves the performance of the algorithm on Dictionary Learning problems.