Multiscale Adaptive Representation of Signals: I. The Basic Framework

Abstract

We introduce a framework for designing multi-scale, adaptive, shift-invariant frames and bi-frames for representing signals. The new framework, called AdaFrame, improves over dictionary learning-based techniques in terms of computational efficiency at inference time. It improves classical multi-scale basis such as wavelet frames in terms of coding efficiency. It provides an attractive alternative to dictionary learning-based techniques for low level signal processing tasks, such as compression and denoising, as well as high level tasks, such as feature extraction for object recognition. Connections with deep convolutional networks are also discussed. In particular, the proposed framework reveals a drawback in the commonly used approach for visualizing the activations of the intermediate layers in convolutional networks, and suggests a natural alternative.

Cited by in corpus (6)

• A Review on Deep Learning in Medical Image Reconstruction
• Towards a Mathematical Understanding of Neural Network-Based Machine Learning: what we know and what we don't
• Adaptive wavelet distillation from neural networks through interpretations
• Learning to Scan: A Deep Reinforcement Learning Approach for Personalized Scanning in CT Imaging
• MetaInv-Net: Meta Inversion Network for Sparse View CT Image Reconstruction
• Wavelet Design in a Learning Framework