Skip to main content
QUICK REVIEW

[论文解读] Supervised Spike Agreement Dependent Plasticity for Fast Local Learning in Spiking Neural Networks

Gouri Lakshmi S, Athira Chandrasekharan|arXiv (Cornell University)|Jan 13, 2026
Advanced Memory and Neural Computing被引用 0
一句话总结

这篇论文引入了Spike Agreement-Dependent Plasticity (SADP)的有监督扩展,用于多层脉冲神经网络的快速局部学习,采用CNN+泊松编码和 Cohen’s kappa 基于更新,在线性时间复杂度内实现竞争性准确性且无需反向传播。

ABSTRACT

Spike-Timing-Dependent Plasticity (STDP) provides a biologically grounded learning rule for spiking neural networks (SNNs), but its reliance on precise spike timing and pairwise updates limits fast learning of weights. We introduce a supervised extension of Spike Agreement-Dependent Plasticity (SADP), which replaces pairwise spike-timing comparisons with population-level agreement metrics such as Cohen's kappa. The proposed learning rule preserves strict synaptic locality, admits linear-time complexity, and enables efficient supervised learning without backpropagation, surrogate gradients, or teacher forcing. We integrate supervised SADP within hybrid CNN-SNN architectures, where convolutional encoders provide compact feature representations that are converted into Poisson spike trains for agreement-driven learning in the SNN. Extensive experiments on MNIST, Fashion-MNIST, CIFAR-10, and biomedical image classification tasks demonstrate competitive performance and fast convergence. Additional analyses show stable performance across broad hyperparameter ranges and compatibility with device-inspired synaptic update dynamics. Together, these results establish supervised SADP as a scalable, biologically grounded, and hardware-aligned learning paradigm for spiking neural networks.

研究动机与目标

  • 为基于SNN的有监督学习提供一种快速、符合生物学的反向传播替代方案的动机。
  • 开发使用总体层面的脉冲一致性度量进行学习的 SADP 有监督扩展。
  • 展示该方法在 CNN 编码预处理和泊松脉冲编码下的可扩展性。
  • 在标准视觉基准和生物医学成像任务上展示性能。
  • 分析在不同数据集和超参数下学习规则的鲁棒性及硬件对齐性。

提出的方法

  • 提出一个有监督的 SADP 规则,其中输出层学习由局部 Hebbian 误差项驱动。
  • 用总体层面的一致性(由 Cohen’s kappa 量化)替代成对脉冲时间点,训练隐藏层。
  • 使用来自原始特征或 CNN 提取特征的泊松编码脉冲作为 SNN 的输入。
  • 实现 1SADP(一个隐藏层)和 2SADP(两个隐藏层)结构,保持严格的局部学习。
  • 通过仅在输出层应用监督并在隐藏层使用基于 kappa 的更新来维持局部性。
  • 通过避免成对脉冲对计算,确保突触更新的线性时间复杂度。
(a) FMNIST
(a) FMNIST

实验结果

研究问题

  • RQ1基于脉冲一致性的有监督局部学习规则(SADP)是否能在不使用反向传播的情况下,在多层 SNNs 中取得有竞争力的准确性?
  • RQ2引入 CNN+泊松编码前端是否提升对复杂和彩色图像的学习和可扩展性?
  • RQ31SADP 与 2SADP 架构在有监督 SADP 下的学习性能和收敛性有何差异?
  • RQ4时域窗口长度(T)和编码方案对不同数据集的准确性与效率有何影响?
  • RQ5有监督的 SADP 方法在数据集、超参数和硬件启发的更新动态下是否鲁棒?

主要发现

  • 在 CNN+泊松编码下,有监督的 SADP 在各数据集上表现出高准确性:MNIST≈99.16%、Fashion-MNIST≈89.9%、CIFAR-10≈70.7%。
  • 仅泊松编码在 MNIST/Fashion-MNIST 上表现良好,但在 CIFAR-10 上显著不及,凸显在复杂数据上需要 CNN 基于特征提取。
  • CNN+泊松编码相比仅泊松编码降低了每轮训练时间(≈18–32 秒),原因是脉冲密度和维度下降。
  • 1SADP 与 2SADP 架构表现相近;在某些数据集上增加第二隐藏层的收益有限。
  • 该方法对严格的局部、无梯度学习规则具有竞争力的结果,并且在使用 CNN+泊松编码时可扩展到生物医学成像任务(如组织病理和 MRI 数据集),表现良好。
Figure 3 : Optimization history plot showing the evolution of validation accuracy across 50 Optuna trials. Blue dots indicate individual trial accuracies, while the orange line represents the cumulative best value. Significant improvements occur around Trials 4, 12–14, and 46–48, reaching a best acc
Figure 3 : Optimization history plot showing the evolution of validation accuracy across 50 Optuna trials. Blue dots indicate individual trial accuracies, while the orange line represents the cumulative best value. Significant improvements occur around Trials 4, 12–14, and 46–48, reaching a best acc

更好的研究,从现在开始

从论文设计到论文写作,大幅缩短您的研究时间。

无需绑定信用卡

本解读由 AI 生成,并经人工编辑审核。