Log-DenseNet: How to Sparsify a DenseNet

TL;DR

This paper introduces Log-DenseNet, a sparsified version of DenseNet that maintains short backpropagation distances with significantly fewer connections, improving scalability and efficiency in deep neural networks.

Contribution

The paper proposes Log-DenseNet, a new connectivity pattern that reduces the number of connections from quadratic to near-linear, enhancing scalability while preserving accuracy.

Findings

Log-DenseNet outperforms DenseNet in semantic segmentation tasks.

Log-DenseNet achieves competitive results in visual recognition.

The proposed method simplifies implementation and scaling of deep networks.

Abstract

Skip connections are increasingly utilized by deep neural networks to improve accuracy and cost-efficiency. In particular, the recent DenseNet is efficient in computation and parameters, and achieves state-of-the-art predictions by directly connecting each feature layer to all previous ones. However, DenseNet's extreme connectivity pattern may hinder its scalability to high depths, and in applications like fully convolutional networks, full DenseNet connections are prohibitively expensive. This work first experimentally shows that one key advantage of skip connections is to have short distances among feature layers during backpropagation. Specifically, using a fixed number of skip connections, the connection patterns with shorter backpropagation distance among layers have more accurate predictions. Following this insight, we propose a connection template, Log-DenseNet, which, in comparison to DenseNet, only slightly increases the backpropagation distances among layers from 1 to ($1 + \log_2 L$), but uses only $L\log_2 L$ total connections instead of $O(L^2)$. Hence, Log-DenseNets are easier than DenseNets to implement and to scale. We demonstrate the effectiveness of our design principle by showing better performance than DenseNets on tabula rasa semantic segmentation, and competitive results on visual recognition.