Structured Sparse Convolutional Autoencoder

TL;DR

This paper introduces a structured sparsity approach in convolutional autoencoders to enhance feature interpretability and prediction accuracy by organizing activations with combined  and  normalization.

Contribution

It proposes a novel structured sparsity function that constrains node activities within feature maps to better capture object structure and shape.

Findings

Improved interpretability of learned features.

Enhanced prediction performance.

Effective organization of activations within and across feature maps.

Abstract

This paper aims to improve the feature learning in Convolutional Networks (Convnet) by capturing the structure of objects. A new sparsity function is imposed on the extracted featuremap to capture the structure and shape of the learned object, extracting interpretable features to improve the prediction performance. The proposed algorithm is based on organizing the activation within and across featuremap by constraining the node activities through $\ell_{2}$ and $\ell_{1}$ normalization in a structured form.