L-CNN: A Lattice cross-fusion strategy for multistream convolutional neural networks

TL;DR

This paper introduces Lattice Cross Fusion, a novel strategy for multistream convolutional neural networks that enhances performance and robustness by crossing signals before pooling layers, demonstrated on CIFAR-10 with significant improvements.

Contribution

The paper presents a new fusion strategy for multistream CNNs that improves accuracy, convergence speed, and robustness over traditional methods.

Findings

Outperforms baseline by 46% on CIFAR-10

Faster convergence and increased stability

Enhanced robustness in image classification

Abstract

This paper proposes a fusion strategy for multistream convolutional networks, the Lattice Cross Fusion. This approach crosses signals from convolution layers performing mathematical operation-based fusions right before pooling layers. Results on a purposely worsened CIFAR-10, a popular image classification data set, with a modified AlexNet-LCNN version show that this novel method outperforms by 46% the baseline single stream network, with faster convergence, stability, and robustness.