We updated SaivDr (Sparsity-Aware Image and Volumetric Data Restoration) package for the first time in about six months.
New this time, we added custom layers and sample codes for use with MATLAB Deep Learning Toolbox. It allows for more flexible DAG configuration than before.
NSOLT enables you to realize Parseval tight, symmetric and multi-resolution convolutional layers, and you can place NSOLT as a convolutional layer in a corner of a convolutional neural network.
We hope you will give it a try.
Acknowledgments: This work was supported by KAKENHI JP19H04135.
The following research topic has been accepted by IEEE/IEIE ICCE-Asia @Seoul, Republic of Korea
Title: ‘Convolutional Nonlinear Dictionary with Cascaded Structure Filter Banks’
Authors: Ruiki Kobayashi and Shogo Muramatsu
The following tutorial is also planed.
Tutorial Title: Sparsity-Aware High-Dimensional Data Restoration with Convolutional Dictionary Learning
Lecturer: Shogo Muramatsu
The event was postponed from April to November due to the impact of COVID-19.
The following work was accepted by IEEE ICASSP2019@Brighton (UK).
Title: ‘CONVOLUTIONAL-SPARSE-CODED DYNAMIC MODE DECOMPOSITION AND ITS APPLICATION TO RIVER STATE ESTIMATION’
Authors: Yuhei Kaneko, Shogo Muramatsu, Hiroyasu Yasuda, Kiyoshi Hayasaka, Yu Otake, Shunsuke Ono, Masahiro Yukawa
Session Type: Poster
Session Title: ‘Enhancement and Restoration II’
Our paper on NSOLT (ICASSP 2014) was cited as a method of structured dictionary learning in the following paper by Prof. Kjersti Engan (University of Stavanger, Norway) famous for the method of optimal directions (MOD).
About NSOLT, please visit here.