An Efficiency-Based Improvement of a Reconstruction Algorithm Reconstructs Signal from Its Scattering Transform

Scattering transform has many advantages and has been widely used in various fields since proposed. An algorithm using machine learning was proposed by some developers of Kymatio to reconstruct the signal based on the Scattering transform of the signal on the internet (Reconstruct a synthetic signal from its scattering transform (DB/OL) and Andreux et al. , 2019). In this paper, I improve this reconstruction algorithm to an efficiency-based algorithm. However, the coefficient in the original formula for the input signal is defined and followed by original gradient descent algorithm with the learning rate multiplied by an accelerator or braker for every iteration. Once the error is spotted out then the system is discarded. By introducing an efficiency parameter, the calculation time and the eliminated error are linked, so that coefficient can be determined to improve the efficiency and eliminated most errors in a short time. Also, the experimenter can adjust the importance parameter in the efficiency parameter to determine the importance of speed and accuracy, which provides the experimenter with more choices.