Learned ISTA with Error-based Thresholding for Adaptive Sparse Coding

Also, it leads to poor generalization to which utilizes a function of the layer-wise reconstruction error test data with a different distribution (or sparsity) from the to suggest a specific threshold for each observation in the training data. To address the above issues, we propose an shrinkage function of each layer. We show that the proposed error-based thresholding (EBT) mechanism of LISTA-based EBT mechanism well disentangles the learnable parameters models to improve their adaptivity. EBT introduces a function in the shrinkage functions from the reconstruction errors, endowing of the evolving estimation error to provide each threshold the obtained models with improved adaptivity to possible in the shrinkage functions in the model. It has no extra learnable data variations. With rigorous analyses, we further show parameter compared with original LISTA-based models, that the proposed EBT also leads to a faster convergence on yet shows significantly better performance.