[论文解读] MeliusNet: Can Binary Neural Networks Achieve MobileNet-level Accuracy?

Joseph Bethge, Christian Bartz+3 more|arXiv (Cornell University)|Jan 16, 2020

Advanced Neural Network Applications参考文献 48被引用 75

一句话总结

MeliusNet 引入了带有 DenseBlock 和 ImprovementBlock 的二值神经网络架构，以提高特征容量和质量，在 ImageNet 上在相似的模型大小和运算量下达到 MobileNet-v1 级别的准确性。

ABSTRACT

Binary Neural Networks (BNNs) are neural networks which use binary weights and activations instead of the typical 32-bit floating point values. They have reduced model sizes and allow for efficient inference on mobile or embedded devices with limited power and computational resources. However, the binarization of weights and activations leads to feature maps of lower quality and lower capacity and thus a drop in accuracy compared to traditional networks. Previous work has increased the number of channels or used multiple binary bases to alleviate these problems. In this paper, we instead present an architectural approach: MeliusNet. It consists of alternating a DenseBlock, which increases the feature capacity, and our proposed ImprovementBlock, which increases the feature quality. Experiments on the ImageNet dataset demonstrate the superior performance of our MeliusNet over a variety of popular binary architectures with regards to both computation savings and accuracy. Furthermore, with our method we trained BNN models, which for the first time can match the accuracy of the popular compact network MobileNet-v1 in terms of model size, number of operations and accuracy. Our code is published online at https://github.com/hpi-xnor/BMXNet-v2

研究动机与目标

• 解决 Binary Neural Networks (BNN) 与紧凑型网络在 ImageNet 上的精度差距。
• 提出一种针对 BNNs 提升特征容量与质量的架构方案。
• 在保持或提升精度的同时，降低残差 32 位层的计算负担。
• 证明 BNNs 在模型大小、FLOPs 和精度方面可以达到 MobileNet-v1–level 的性能。

提出的方法

• 引入 MeliusNet，交替使用 DenseBlock（提升特征容量）与 ImprovementBlock（提升特征质量）。
• 每个 DenseBlock 使用 64 个新的二值通道，拼接到特征图后，再接一个 64 通道的二值 ImprovementBlock，并带有残差连接以细化这些特征。
• 用分组干线方法重新设计前几层（用三个分组的 3x3 卷积替代 7x7 卷积），在不增加模型大小的情况下大幅降低 32-bit 运算。
• 避免对初始、最终和部分下采样卷积进行二值化以保持准确性，并基于实证结果省略通道级缩放因子。
• 在 ImageNet 上从头训练模型，使用基于 STE 的二值化，配以 Adam/RAdam 优化器和余弦学习率调度，包括可选的 warm-up。

实验结果

研究问题

• RQ1 Can a binary neural network architecture be designed to match MobileNet-v1 in accuracy at comparable size and compute on ImageNet?
• RQ2 Do architectural innovations tailored for BNNs (capacity-increasing blocks and quality-improving blocks) yield measurable gains over prior binary architectures?
• RQ3 Does a grouped-stem 32-bit layer design reduce operations enough to close the gap with compact networks without sacrificing accuracy?
• RQ4 Are improvements in BNN performance specific to binary architectures, or do they extend to 32-bit networks as well?

主要发现

• MeliusNet 在 ImageNet 上，在相似的模型尺寸和运算量下，一直优于以往的二值架构。
• 分组干线在测试模型中显著降低了 32-bit 运算，同时提升了准确性。
• 32-bit 版本的 MeliusNet 未显示出对 DenseNet 的改进，表明这些好处特定于 BNNs。
• MeliusNet 的配置可以在多种目标尺寸/运算量下达到 MobileNet-v1 的准确性，显示出与紧凑网络的竞争力。
• 所提分组干线结合 MeliusNet 架构缩小了 BNNs 与 MobileNet-v1 之间的准确性差距，凸显在合适硬件上的潜在能耗节省。