[论文解读] Spiking-YOLO: Spiking Neural Network for Energy-Efficient Object Detection

Seijoon Kim, Seongsik Park+2 more|arXiv (Cornell University)|Mar 12, 2019

Advanced Memory and Neural Computing参考文献 40被引用 49

一句话总结

Spiking-YOLO 引入通道级归一化和带不平衡阈值的有符号神经元，以实现深度SNN对象检测，在 VOC 和 COCO 上达到接近 Tiny YOLO 的精度，但能耗大幅降低。

ABSTRACT

Over the past decade, deep neural networks (DNNs) have demonstrated remarkable performance in a variety of applications. As we try to solve more advanced problems, increasing demands for computing and power resources has become inevitable. Spiking neural networks (SNNs) have attracted widespread interest as the third-generation of neural networks due to their event-driven and low-powered nature. SNNs, however, are difficult to train, mainly owing to their complex dynamics of neurons and non-differentiable spike operations. Furthermore, their applications have been limited to relatively simple tasks such as image classification. In this study, we investigate the performance degradation of SNNs in a more challenging regression problem (i.e., object detection). Through our in-depth analysis, we introduce two novel methods: channel-wise normalization and signed neuron with imbalanced threshold, both of which provide fast and accurate information transmission for deep SNNs. Consequently, we present a first spiked-based object detection model, called Spiking-YOLO. Our experiments show that Spiking-YOLO achieves remarkable results that are comparable (up to 98%) to those of Tiny YOLO on non-trivial datasets, PASCAL VOC and MS COCO. Furthermore, Spiking-YOLO on a neuromorphic chip consumes approximately 280 times less energy than Tiny YOLO and converges 2.3 to 4 times faster than previous SNN conversion methods.

研究动机与目标

• 通过利用 SNNs，推动对象检测在超越图像分类的能效提升。
• 解决深SNN在回归任务（边界框坐标）中的训练与信息传输挑战。
• 开发方法使深SNN中的基于脉冲的对象检测更准确并评估能效。

提出的方法

• 提出通道级归一化，以防止过低激活并提高深SNN的放电率。
• 引入带不平衡阈值的有符号神经元（IBT），以在SNN中真实实现 leaky-ReLU 漏泄项。
• 将 DNN-to-SNN 转换方法应用于对象检测，以 TinyYOLO 作为基础网络。
• 在 SNN 框架内实现最大池化和批量归一化。
• 评估解码方案：膜电位解码与脉冲计数解码，膜电位解码提供更高的精度。
• 使用 TensorFlow Eager 在 NVIDIA V100 GPU 上进行仿真，并与 Tiny YOLO 和神经形态硬件进行比较。

实验结果

研究问题

• RQ1 深度SNN 能否在非平凡数据集（VOC 和 COCO）上训练出与 DNN 竞争的对象检测精度？
• RQ2 通道级归一化和 IBT 启用的有符号神经元是否能克服深SNN在回归任务中的归一化与 leaky-ReLU 实现问题？
• RQ3 Spiking-YOLO 相对于 Tiny YOLO，在 GPU 与神经形态硬件上的能效权衡如何？
• RQ4 哪种输出解码方案能为基于脉冲的对象检测提供更高的精度？

主要发现

• Spiking-YOLO 在特定配置下，在 VOC 和 COCO 数据集达到最高约 Tiny YOLO 性能的 98%。
• 通道级归一化显著提高跨通道的放电率并加速收敛（比层归一化更快）。
• 带不平衡阈值的有符号神经元在 SNN 中有效实现 leaky-ReLU，对检测性能至关重要。
• 基于膜电位的解码在准确性和收敛速度方面优于脉冲计数解码。
• 在 32-bit 浮点或整数 MAC/AC 模型上，Spiking-YOLO 的能耗比 Tiny YOLO 高效超过 2000 倍，在神经形态 TrueNorth 硬件上约节能 280 倍。