The winner-take-all (WTA) mechanism based on the inhibitory dynamics of vertical-cavity surface-emitting laser with an embedded saturable absorber (VCSEL-SA) neurons is proposed for the first time. The WTA mechanism is shown numerically in a photonic spiking neural network (SNN). The effect of bias current on WTA time window is analyzed based on the proposed SNN. Moreover, a pattern recognition approach is presented numerically based on the WTA mechanism in all-optical neural network consisting of VCSEL-SA neurons. For the pattern recognition, the robustness of noisy inputs is examined. The effects of bias current, strength of inhibition and weight matrix on the speed of pattern recognition are analyzed carefully. Furthermore, the max-pooling operation is implemented numerically in the all-optical VCSEL-SA neural network model based on the WTA mechanism for the first time. The results hold great promise for the development of energy-efficient and high-speed photonic spiking neural network.

中文翻译：

---

模式识别和最大池化操作的全光系统的赢家通吃机制

首次提出了基于具有嵌入式饱和吸收体（VCSEL-SA）神经元的垂直腔面发射激光器的抑制动力学的赢家通吃（WTA）机制。WTA 机制在光子尖峰神经网络 (SNN) 中以数字方式显示。基于所提出的 SNN 分析偏置电流对 WTA 时间窗口的影响。此外，在由VCSEL-SA神经元组成的全光神经网络中，基于WTA机制在数值上提出了一种模式识别方法。对于模式识别，检查噪声输入的鲁棒性。仔细分析了偏置电流、抑制强度和权重矩阵对模式识别速度的影响。此外，首次在基于 WTA 机制的全光 VCSEL-SA 神经网络模型中数值实现了最大池化操作。结果为开发节能和高速光子尖峰神经网络提供了广阔的前景。