Understanding Geometry of Encoder-Decoder CNNs

Abstract

Encoder-decoder networks using convolutional neural network (CNN) architecture have been extensively used in deep learning literatures thanks to its excellent performance for various inverse problems. However, it is still difficult to obtain coherent geometric view why such an architecture gives the desired performance. Inspired by recent theoretical understanding on generalizability, expressivity and optimization landscape of neural networks, as well as the theory of convolutional framelets, here we provide a unified theoretical framework that leads to a better understanding of geometry of encoder-decoder CNNs. Our unified mathematical framework shows that encoder-decoder CNN architecture is closely related to nonlinear basis representation using combinatorial convolution frames, whose expressibility increases exponentially with the network depth. We also demonstrate the importance of skipped connection in terms of expressibility, and optimization landscape.

Accepted to ICML 2019

Cited by in corpus (8)

• EEGFuseNet: Hybrid Unsupervised Deep Feature Characterization and Fusion for High-Dimensional EEG with An Application to Emotion Recognition
• Learning Dynamic Graph Representation of Brain Connectome with Spatio-Temporal Attention
• Geometric Approaches to Increase the Expressivity of Deep Neural Networks for MR Reconstruction
• k-Space Deep Learning for Accelerated MRI
• Understanding Graph Isomorphism Network for rs-fMRI Functional Connectivity Analysis
• Boosting CNN beyond Label in Inverse Problems
• Differentiated Backprojection Domain Deep Learning for Conebeam Artifact Removal
• Attention-gated convolutional neural networks for off-resonance correction of spiral real-time MRI