Knowing the connectivity patterns in neural circuitry is essential to understand the operating mechanism of the brain, as it allows the analysis of how neural signals are processed and flown through the neural system. With the recent advances in neural recording technologies in terms of channel size and time resolution, a simple and efficient system to perform neural connectivity inference is highly desired, which will enable the process of high dimensional neural activity recording data and reduction of the computational time and cost. In this work, we show that the spike-timing dependent plasticity (STDP) algorithm can be used to reconstruct neural connectivity patterns in a biological neural network, with higher accuracy and efficiency than statistic-based inference methods. The biologically inspired STDP learning rules are natively implemented in a second-order memristor network and are used to estimate the type and the direction of neural connections. When stimulated by the recorded neural spike trains, the memristor device conductance is modulated by the proposed STDP learning rules, which in turn reflects the correlation of the spikes and the possibility of neural connections. By compensating for the different levels of neural activity, highly reliable inference performance can be achieved. The proposed approach offers real-time and local learning, resulting in reduced computational cost/time and strong tolerance to variations of the neural system.

知道神经电路中的连通性模式对于理解大脑的运作机制至关重要，因为它可以分析神经信号如何被处理和流过神经系统。随着神经记录技术在通道大小和时间分辨率方面的最新进展，迫切需要一种简单有效的系统来执行神经连接推理，这将使高维神经活动记录数据的过程得以实现，并减少了计算时间和成本。在这项工作中，我们表明，与基于统计的推理方法相比，峰值定时依赖可塑性（STDP）算法可用于重建生物神经网络中的神经连通性模式。受生物启发的STDP学习规则在二阶忆阻器网络中本地实现，并用于估计神经连接的类型和方向。当受到记录的神经尖峰序列刺激时，忆阻器器件电导由拟议的STDP学习规则进行调节，这反过来又反映了尖峰的相关性和神经连接的可能性。通过补偿神经活动的不同水平，可以实现高度可靠的推理性能。所提出的方法提供了实时和本地学习，从而减少了计算成本/时间，并且对神经系统的变化具有很强的容忍性。忆阻器器件电导由拟议的STDP学习规则进行调节，这反过来又反映了尖峰的相关性和神经连接的可能性。通过补偿神经活动的不同水平，可以实现高度可靠的推理性能。所提出的方法提供了实时和本地学习，从而减少了计算成本/时间，并且对神经系统的变化具有很强的容忍性。忆阻器器件电导由拟议的STDP学习规则进行调节，这反过来又反映了尖峰的相关性和神经连接的可能性。通过补偿神经活动的不同水平，可以实现高度可靠的推理性能。所提出的方法提供了实时和本地学习，从而减少了计算成本/时间，并且对神经系统的变化具有很强的容忍性。